Aβ can increase gene expression and protein production of a number of cytokines by human microglia in vitro, including interleukin-1 beta (IL1β), IL8, IL10, IL12, tumour necrosis factor alpha (TNFα), macrophage inflammatory protein-1 alpha (MIP1α, aka CCL3), MIP1β (CCL4) and monocyte chemoattractant protein-1 (MCP1, CCL2) (Lee et al., 2002). It is no surprise, therefore, that their expression is reported to be altered in the brain, blood and cerebrospinal fluid (CSF) of AD patients (Akiyama et al., 2000). Animal models of AD, such as the APP transgenic line Tg2567 carrying the Swedish mutation, also show enhanced levels of TNFα, IL1β, IL1α, chemoattractant protein-1, COX2 and complement component 1q (Sastre et al., 2008).

Despite the strong links between inflammation and AD, there are a surprisingly small number of studies assessing the expression of many inflammatory cytokines. Evidence of an increased risk of AD is observed when the IL10−1082 GA gene polymorphism is present; however, the association is only marginally significant (Di Bona et al., 2012). Adeno-associated virus (AAV)-mediated gene delivery of this anti-inflammatory cytokine reduced glial activation, enhanced Aβ clearance into the plasma, upregulated neurogenesis and improved spatial memory in APP–PS1 mice (Kiyota et al., 2012); however, it had no effect on Aβ plaque load. Increased concentrations of p40 (a common subunit of the pro-inflammatory cytokines IL12 and IL23) are found in the CSF of AD patients, and this correlates with cognitive performance (vom Berg et al., 2012). In addition, ablation of IL12 and IL23 signalling in APP–PS1 mice reduces amyloid burden (vom Berg et al., 2012). IL6 is elevated in the serum of patients with AD (Helmy et al., 2012), and there is an interaction between IL6 and IL10 gene polymorphisms associated with increased risk for AD (Arosio et al., 2004; Combarros et al., 2009). These studies strongly suggest that cytokine dysregulation is a significant candidate for initiating the development of AD. The most commonly studied cytokines are IL1β and TNFα; up-regulation of these pro-inflammatory cytokines has been heavily implicated as common cellular response markers to Aβ release and deposition (Hickman et al., 2008).

IL1 is a primary mediator of the pro-inflammatory response in the body and is capable of inducing the production of many other cytokines. A number of studies have analysed the possibility that polymorphisms in IL1 could lead to increased risk for the development of AD. A specific IL1α gene polymorphism in allele 2 triples the risk of developing AD (Du et al., 2000; Grimaldi et al., 2000; Rebeck, 2000), and an even greater risk is seen with patients carrying both IL1α and IL1β allele 2 gene polymorphisms (Griffin et al., 2000). In addition, this IL1α gene polymorphism is associated with increased microglial cell numbers in post-mortem AD brains, and the increase was even higher in patients carrying the APOE ϵ4 allele (Hayes et al., 2004). These polymorphisms increase expression of IL1 and may therefore drive IL1-mediated mechanisms that lead to an overall enhancement of neuroinflammation in the brain and consequently increased risk for AD. Overexpression of IL1β has been reported in microglia and astrocytes of AD brains since the late 1980s (Griffin et al., 1989), in particular in plaque-associated microglia (Mrak and Griffin, 2005). This association is transient; high IL1β expression is associated with early amyloid deposits and the conversion of diffuse amyloid plaques into compact forms (Griffin et al., 1995). IL1α and IL1β increase translation of APP in astrocytes in vitro, with the latter enhancing synthesis to a greater extent (Rogers et al., 1999). In addition, IL1β drives astrocyte activation and increases β-secretase-cleaved carboxyl-terminal fragment (βCTF) production in vivo, suggesting that this cytokine may have a significant role to play in the progression of AD (Sheng et al., 1996). Indeed, many argue that early and sustained overexpression of IL1 could be an initiator of disease pathology (Griffin and Mrak, 2002). IL1 may also contribute directly to neuronal dysfunction in AD; chronic release of exogenous IL1β in vivo induced overexpression and phosphorylation of neurofilament protein and tau, which are components of neurofibrillary tangles of AD (Sheng et al., 2000). This overexpression was specifically associated with swollen, dystrophic neurites. Interestingly, a recent paper by Ghosh and colleagues reports that sustained IL1β overexpression in a triple transgenic (3xTg) mouse model of AD increases tau hyperphosphorylation but leads to a corresponding decrease in amyloid burden (Ghosh et al., 2013). They also found a four- to sixfold increase in plaque-associated microglia, suggesting that IL1β may directly affect tau phosphorylation, possibly via p38 mitogen-activated kinase and glycogen synthase kinase-3β activity, but regulates amyloid indirectly via glial activation. This builds on similar findings from their group showing that overexpression of IL1β specific to the hippocampus in amyloid precursor protein–presenilin-1 (APP–PS1) mice induces significant astrocyte and glial activation with a corresponding decrease in amyloid plaque deposition (Shaftel et al., 2007). The differential effect of IL1 on tau and amyloid makes it an unlikely candidate for therapeutic manipulation; however, it does lead to intriguing questions about targeting certain microglial or astrocytic processes to reduce amyloid burden.

TNFα is increased in the serum of AD patients (Fillit et al., 1991; Bruunsgaard et al., 1999; Tarkowski et al., 1999), and increased expression is found in activated microglia surrounding amyloid plaques in brains of AD patients (Dickson et al., 1993). Microglia produce TNFα that is neurotoxic in response to Aβ exposure (Meda et al., 1995; Combs et al., 2001), and Aβ itself can induce neuronal apoptosis via TNF receptor type I (TNF-RI) (Li et al., 2004). These studies indicate that there may be TNFα-mediated apoptosis in AD. Increased TNFα is also found in 3xTg-AD and APPswe mouse models, expressed by both neurons and microglia (Mehlhorn et al., 2000; Janelsins et al., 2005). Overexpression of TNFα in 3xTg-AD mice induces significant neurodegeneration (Janelsins et al., 2008), providing further evidence for its role in the cell death seen in AD. However, chronic down-regulation of TNF-RII in 3xTg-AD mice exacerbates amyloid and tau pathology (Montgomery et al., 2013), although interestingly this effect was evident only in mice whose TNF-RII receptors were knocked down from 12 to 21 months of age. Young and aged mice in which TNF-RI or both TNF-RI and TNF-RII receptors were knocked down showed no change in amyloid burden or tau hyperphosphorylation. TNF-RI and TNF-RII show divergent signalling pathways, with TNF-RI being heavily implicated in neuronal apoptosis (Yang et al., 2002), whereas TNF-RII is reported to activate alternative nuclear factor kappa B (NF-κB) signalling associated with the down-regulation of pro-inflammatory cytokines (Rauert et al., 2010). This once again highlights the complex nature of cytokine involvement in AD, and any therapeutic intervention targeted at TNFα must take into account disease state and receptor specificity.

Pro-inflammatory cytokines are known to affect Aβ generation, suggesting that a positive feedback loop may develop following an Aβ-triggered inflammatory response in which cytokines contribute to further Aβ production (Sastre et al., 2006a,b). The potential mechanisms include increasing the susceptibility for Aβ deposition or aggregation (Games et al., 1995; Guo et al., 2002), up-regulating β-secretase (BACE1) mRNA (Sastre et al., 2003, 2006b) and elevating APP transcription (Amara et al., 1999; Rogers et al., 1999).

ROS are chemically reactive molecules that are cytotoxic by-products of oxygen metabolism. These include hydroxyl radicals (−OH), peroxide (H₂O₂) and superoxide radicals (O₂⁻). Endogenous ROS are mainly produced by mitochondria, peroxisomes, endoplasmic reticulum and the NADPH oxidative complex (NOX) in the cell membranes (Vitale et al., 2013). The main mechanisms for eliminating ROS are superoxide dismutases (SODs), catalase, heme-oxygenase and glutathione peroxidase, enzymes that are increased in the vicinity of Aβ plaques (Furuta et al., 1995; Pappolla et al., 1998). During ageing or inflammation, there is accumulation of ROS, which causes oxidation of lipids, sugars, proteins and DNA, thus leading to cell dysfunction and consequently cell death (Chami and Checler, 2012). The neurons are highly susceptible to oxidative stress because of their high metabolic demands and poor anti-oxidant capabilities. ROS can also contribute to further inflammation through induction of cyclooxygenase-2 (COX2), inflammatory cytokines and pro-inflammatory transcription factors such as NF-κβ (Yan et al., 1995). In AD brains, several oxidative markers are increased (Guglielmotto et al., 2010), including protein oxidation (Smith et al., 1997), lipid peroxidation (Mark et al., 1997; Sayre et al., 1997), advanced glycation end products (Smith et al., 1994) and oxidation of nucleic acids (Nunomura et al., 1999).

Oxidative stress is greatest early in the disease (Nunomura, 2001). Increased levels of isoprostanes, a marker for lipid peroxidation, were detected in the urine, blood and CSF of both mild cognitive impairment (MCI) and AD patients (Praticò et al., 2000, 2002; Mecocci, 2004), suggesting that oxidative markers could be potentially used as biomarkers for the early detection of AD. Increased ROS and oxidative stress were also confirmed in mouse models of AD (Pappolla et al., 1998; Smith et al., 1998; Matsuoka et al., 2001; Praticò et al., 2001).

It is well established that Aβ causes oxidative stress which in turn causes more Aβ, resulting in a positive feedback loop that further contributes to the progress of the disease (Tamagno et al., 2012; Butterfield et al., 2013). This is because ROS affect the rate-limiting enzyme of AD, beta-site APP cleaving enzyme-1 (BACE1), by altering its distribution in non-apoptotic conditions (Tan et al., 2013) as well as its activity and expression (Tamagno et al., 2002). Oxidative stress-mediated BACE activity is γ-secretase-dependent as absence of PS1 abolished the effect (Tamagno et al., 2008). Further studies strengthened the correlation of BACE1, γ-secretase and oxidative stress, showing that treatment with dehydroepiandrosterone, an antioxidant, reduced BACE1 activity in NT2 neurons exposed to oxidative stress (Tamagno et al., 2003) and that mild impairment of oxidative metabolism by thiamine deficiency increased Aβ, carboxyl-terminal fragments and plaque burden in an AD mouse model (Karuppagounder et al., 2009).

Dysfunctional homeostasis of transition metals, such as copper and particularly iron (Bush, 2013), may play an important role in the pathogenesis of AD by enhancing oxidative stress. (Crichton et al. 2011). Increased concentrations of iron are present in the basal ganglia (Bartzokis and Tishler, 2000), hippocampus (Ding et al., 2009) and neocortex (Cornett et al., 1998) and in or around the plaques and tangles (Morris et al., 1994). In addition, iron modulates the ability of α-secretase to cleave APP (Rivera-Mancía et al., 2010), promotes Aβ toxicity (Bodovitz et al., 1995) and aggregation, and directly regulates the expression and the synthesis of APP via the iron responsive element at the 5′ untranslated region of APP mRNA (Rogers et al., 2002; Rivera-Mancía et al., 2010). Levels of furin, an enzyme responsible for the activation of β-secretases involved in the amyloidogenic pathway, are also modulated by iron (Crichton et al., 2011), which will play an important role in AD pathogenesis (Silvestri and Camaschella, 2008).

Complement receptors are one of the categories of cell surface molecules on microglia that are up-regulated in response to the activation of these cells (Liu and Hong, 2003). Aβ-induced complement activation leads to generation of C1q, C4 and C3 activation fragments around the plaques. Here microglia express complement proteins C1q and C3 and receptors C1qR, CR3, CR4 and C5aR, which support phagocytic uptake (Keene et al., 2011). Inhibition of the complement system results in an increase of Aβ plaque formation and neurodegeneration in AD transgenic mice (Shen and Meri, 2003).

In contrast, lack of C1q in mouse models of AD results in decreased pathology (Hafer-Macko et al., 2000). This indicates that one mechanism by which microglia could recruit further reactive cells to the site of a plaque and cause neurotoxic damage is by activating the classical complement pathway and the inflammatory machinery associated with it (i.e. pro-inflammatory cytokines and oxidative products) through production of C1q (McGeer and McGeer, 1998; Bonifati and Kishore, 2007).

There have been a number of reports demonstrating the presence of T cells in the brain of AD patients since the original observation 25 years ago (Rogers et al., 1988). Furthermore, inflammatory IFNγ-secreting Th1 cells and IL17–secreting Th17 cells have been shown to infiltrate the brain of older APP–PS1 mice (Browne et al., 2013).

Therapeutic vaccination with Aβ antibodies in mice evidenced the Fc-mediated uptake and clearance of Aβ antibody complexes by local activated microglia (Bard et al., 2000). However, because a human trial of Aβ immunization led to meningoencephalitis in some patients, this treatment has been discontinued. Recently, it was found that nasal vaccination in mice was able to decrease Aβ, and the extent of this reduction correlated with microglial activation, suggesting that this may be a promising approach for human Aβ immunization (Frenkel et al., 2005).

The finding that treatment with non-steroidal anti-inflammatory drugs (NSAIDs) is associated with a reduced risk and age of onset of AD reinforced the hypothesis that modulating inflammation could have therapeutic efficacy. The beneficial effects of NSAIDs have also been associated with reductions in Aβ generation, because experiments in vitro and in AD animal models indicate that certain NSAIDs are able to decrease Aβ levels, plaque size and tau phosphorylation (Yoshiyama et al., 2007; El Khoury and Luster, 2008; Sastre and Gentleman, 2010).

However, clinical trials have failed to reproduce the beneficial effects of NSAIDs in AD patients. This has led to further analysis of the previously published epidemiological data, which has revealed that the use of NSAIDs prevents cognitive decline in older adults if started in midlife (prior to age 65) rather than late in life (Hayden et al., 2007). In addition, it was recently shown that while NSAIDs may indeed protect those with healthier brains, they can accelerate AD pathogenesis in patients with advanced stages of the disease (Breitner et al., 2009). This is supported by studies in transgenic mice, in which NSAIDs can prevent the appearance of cell cycle protein markers in neurons in young mice, but not after cell cycle entry has been initiated (Varvel et al., 2009). Therefore, it seems that the protective effects of NSAIDs depend very much on the stage of the disease at which the medication is started as well as the duration of the treatment (Sastre and Gentleman, 2010).

PPARγ inhibition regulates the transcription of pro-inflammatory genes, such as IL1β; therefore, activation of PPARγ consequently inhibits the inflammatory response. In addition, our group found that PPARγ activators are able to decrease total Aβ levels under inflammatory conditions by affecting BACE1 transcription (Sastre et al., 2003, 2006b). Recently it was shown that ibuprofen is able to suppress RhoA activity in neuronal cells through PPARγ activation, promoting neurite elongation (Dill et al., 2010). Therefore, PPARγ activation could be beneficial in AD at several levels.

However, other groups have suggested that PPARγ may affect Aβ clearance and degradation. Landreth demonstrated that PPARγ activation induces lxrα, apoe, and abca1 expression, promoting Aβ clearance by both microglia and astrocytes (Mandrekar-Colucci et al., 2012). Furthermore, PPARγ stimulates an increase of Aβ phagocytosis, which was mediated by the up-regulation of scavenger receptor CD36 expression. In addition, combined treatment with agonists for the heterodimeric binding partners of PPARγ, the retinoid X receptors (RXRs), showed additive enhancement of the Aβ uptake that was mediated by RXRα activation (Yamanaka et al., 2012).

A recent prospective randomized, open-controlled study with pioglitazone (a typical PPARγ agonist) showed that at 6 months, the Wechsler Memory Scale-Revised logical memory-I scores significantly increased in the pioglitazone group, but not in the control group (Hanyu et al., 2009). Another PPARγ agonist, rosiglitazone, has been trialled with inconsistent results. In contrast to pioglitazone, rosiglitazone cannot cross the blood–brain barrier (BBB) (Festuccia et al., 2008), and it was suggested that the protective effects are mediated through its effects on insulin and glucocorticoids that are able to penetrate the brain.

Minocycline, a tetracycline derivative, has potent anti-inflammatory, anti-apoptotic and neuroprotective properties. Minocycline is able to cross the BBB and effectively delays disease progression and reduces neuronal death in mouse models of amyotrophic lateral sclerosis (Zhu et al., 2002), Huntington’s disease (Chen et al., 2000) and Parkinson’s disease (Wu et al., 2002). In many cases, the neuroprotective properties of minocycline have been attributed to inhibition of caspases. In primary cortical neurons, minocycline was shown to reduce caspase-3 activation and lowered the generation of caspase-3-cleaved tau fragments (Noble et al., 2009). Recently, minocycline was shown to protect against Aβ-induced cell death and prevent fibrillization of Aβ in vitro (Familian et al., 2006), reduce iNOS levels (Ferretti et al., 2012), prevent Aβ deposition and cognitive decline in APP transgenic mice (Seabrook et al., 2006; Ferretti et al., 2012) by reducing BACE1 levels (Ferretti et al., 2012) and inhibit neuronal death and attenuate learning and memory deficits following administration of Aβ to rats (Ryu et al., 2004; Choi et al., 2007). In addition, treatment of tangle-forming transgenic mice (htau line) with minocycline resulted in reduced levels of tau phosphorylation and insoluble tau aggregates (Noble et al., 2009).

Another potential mechanism of action of minocycline has been related to the inhibition of microglial activation. Administration of minocycline in animal models of ALS attenuated the induction of the expression of M1 microglia markers during the progressive phase, whereas it did not affect the transient enhancement of expression of M2 microglia markers during the early pathogenesis phase (Kobayashi et al., 2013). This study suggests that minocycline may selectively inhibit the microglia polarization to a pro-inflammatory state.

Interestingly, anti-TNFα treatment reduced Aβ and tau phosphorylation in transgenic mice. Treatment with the antibody against TNFα, infliximab, increased the number of CD11c-positive dendritic-like cells and the expression of CD11c. These data suggested that the CD11c-positive dendritic-like cells might contribute to the infliximab-induced reduction of AD-like pathology (Shi et al., 2011).