Inflammation has repeatedly been implicated in AD but the source for this inflammation has been elusive. Inflammatory proteins in blood, notably C reactive protein (CRP) and IL6, are elevated several years before the clinical onset of dementia in several studies (Schmidt et al., 2002; Engelhart et al., 2004; Tilvis et al., 2004; Kuo et al., 2005). A high plasma CRP has been associated with a 3-fold increased risk of developing AD years later (Schmidt et al., 2002). Another study showed cognitively intact older individuals in the top tertile for leukocyte IL1β or TNFα production have ~3 times increased risk of developing AD compared to those in the lowest tertile (Tan et al., 2007). Though clinical trials have shown non-steroidal anti-inflammatory drugs (NSAIDs) do not affect cognitive decline in AD (de Craen et al., 2005; Imbimbo, 2009; Imbimbo et al., 2010), meta-analyses show regular NSAID use is associated with a 2 fold reduction in the odds of developing AD (McGeer et al., 1996). Moreover, a large case control study (>200,000 subjects) showed that the regular use of NSAIDs reduced the risk of developing AD (Vlad et al., 2008). Thus, NSAIDS may delay the onset of AD, but once it develops NSAIDS do not appear to affect the course of AD, suggesting an opportunity for intervention prior to disease onset (Grammas, 2011; Butchart and Holmes, 2012). These data suggest that inflammation is occurring for some time prior to onset of AD pathology and symptoms, but there has been little indication of what drives the inflammation that goes on for years. This review proposes this may be due to Gram-negative bacterial LPS in blood and in brain of AD subjects.

Delirium, which is often caused by infection, is associated with increased incidence of subsequent development of dementia in cognitively intact old individuals (Rahkonen et al., 2000a,b). The presence of one or more infections over a 5-year follow up period increased the odds of developing AD, and risk increased with age (Dunn et al., 2005). Receiving DPT vaccines early in life as well as other vaccines later in life significantly reduces the risk of subsequent AD (Tyas et al., 2001; Verreault et al., 2001). Protection by DPT vaccine may be due in part to preventing infection by the Gram-negative Bordetella Pertussis bacterium which causes whooping cough. Tooth loss (Stein et al., 2007) and oral infections (Poole et al., 2013; Abbayya et al., 2015; Chen et al., 2017) have been associated with AD. Furthermore, Porphyromonas gingivalis (Ishida et al., 2017) and LPS from Porphyromonas gingivalis (Wu et al., 2017) produce AD-like phenotypes in mice. In addition, Spirochetes (Miklossy, 1993; Miklossy et al., 1994, 2004; Riviere et al., 2002), chlamydophila pneumonia (Hammond et al., 2010), Helicobacter pylori (Kountouras et al., 2009), fungi (Pisa et al., 2015), herpes viruses (Civitelli et al., 2015) and cytomegalovirus (Lovheim et al., 2018) are also reported to be involved in with AD pathology. Interestingly, previous in vitro studies demonstrate that amyloid-like morphological changes occur following Borrelia burgdorferi spirochetes or LPS exposure. These studies suggest that AD might be a neurological disorder associated with infectious agents.

It is possible that several different infections might initiate downstream AD pathology. The possibility of infection received a significant boost with the recent discovery that every human brain examined had evidence of Gram-negative bacteria (Branton et al., 2013; Emery et al., 2017) though the issue of contamination in these studies has yet to be resolved. For this review the focus is on Gram-negative bacterial LPS as it is associated with and may cause AD pathology.

There is considerable evidence for vascular abnormalities in AD. Cerebral blood flow is reduced in AD prior to cognitive decline (Ruitenberg et al., 2005). Patients with an APOE4 allele have disrupted fMRI connectivity (blood flow correlations) in the absence of amyloid plaques (detected by PET) or decreased CSF Aβ₄₂ (Sheline et al., 2010). Cerebral glucose metabolism is decreased in preclinical and prodromal AD before symptoms (Hunt et al., 2007; Herholz, 2010). Microvessels isolated from AD patients release many cytokines, chemokines and proteases compared to control patients (Grammas, 2011). In genetic mouse models of AD, alterations of blood flow and BBB permeability occur before symptoms and before amyloid beta deposition (Iadecola et al., 1999; Ujiie et al., 2003; Iadecola, 2004). Using vascular corrosion casts, the 3D arrangement of the brain vessels is abnormal in a mouse model of AD prior to appearance of amyloid plaques (Meyer et al., 2008). In addition, BBB breakdown has been demonstrated in humans with mild cognitive impairment (MCI) and early AD before brain atrophy and dementia (Montagne et al., 2017). These data provide strong evidence for neurovascular abnormalities in sporadic AD in humans and in mouse genetic AD models prior to appearance of brain amyloid plaque neuropathology (Zlokovic, 2005, 2008, 2011) and support the fact that both cardiovascular disease and cerebrovascular disease are significant risk factors for sporadic AD (Tyas et al., 2001; Veurink et al., 2003; White et al., 2005; Zhu et al., 2007; Marlatt et al., 2008; de la Torre, 2009; Guglielmotto et al., 2009). Notably, cerebral vascular disease and AD pathology co-exist in up to 80% of aging human brains (Schneider et al., 2007; Savva et al., 2009; Iadecola, 2013). These data are relevant for our finding of LPS in brain described below, since areas of ischemic and/or hypoxic injury might provide a portal for LPS entry from blood into human brain.

Recent neuroimaging studies in individuals with MCI and early AD have shown BBB breakdown in the hippocampus (Montagne et al., 2015) and several Gray Matter (GM) and White Matter (WM) regions (van de Haar et al., 2016a,b, 2017).

Besides the original discovery of the association of Apolipoprotein E4 with sporadic AD (Bertram et al., 2010; Bertram and Tanzi, 2012; Goate and Hardy, 2012), additional genetic insights into sporadic AD have been made using GWAS (Bertram et al., 2010; Bertram and Tanzi, 2012). The genes implicated in these studies are involved in Aβ clearance at the blood brain barrier (APOE, BIN1, CLU, CR1, PICALM; Zlokovic et al., 1996; O’Brien and Wong, 2011; Wu et al., 2012), in inflammation in blood and brain (APOE, CD33, CLU, CR1, HLA-DRB5/1, INPP5D, TREM2; Bertram et al., 2010; Veerhuis, 2011; Bertram and Tanzi, 2012; Crehan et al., 2012), in the immune response (CR1, CD33, MS4A, CLU, ABCA7, EPHA1, HLA-DRB5-HLA-DRB1), endocytosis (BIN1, PICALM, CD2AP, EPHA1 and SORL1) and lipid biology (CLU, ABCA7 and SORL1; Karch and Goate, 2015). Many of these genes are expressed by peripheral monocytes and brain microglia (Villegas-Llerena et al., 2016) and could be associated with an infectious cause of sporadic AD (Heneka et al., 2015). Our model of how LPS could lead to AD neuropathology could incorporate all these genes and molecules (see below). Since Aβ also acts on TLR4 receptors on both monocytes and neutrophils, this could help explain how neutrophils promote AD-like pathology with cognitive decline via the leukocyte LFA-1 adhesion molecule in two AD mouse models (Zenaro et al., 2015).

Myelin injury has been recognized in AD brain for some time, including the very first report of AD pathology by Alzheimer (Alzheimer et al., 1991; Scheltens et al., 1995; Englund, 1998; Möller and Graeber, 1998; Bartzokis, 2011). The relationship of the myelin injury to the other better-known neuropathology of AD, including amyloid plaques and tau-neurofibrillary tangles, has been unclear and the explanation for such myelin injury has been equally obscure. However, recently myelin injury has been suggested as an important/principal component of AD pathophysiology. The volume of White Matter Hyperintensities (WMH) in AD brain predicts the rate and severity of cognitive decline (Brickman et al., 2008). The Low-density lipoprotein receptor-related protein 1, which transports Aβ out of brain, is also an essential receptor for myelin phagocytosis providing a link between myelin damage and Aβ (Gaultier et al., 2009). Moreover, Aβ_1–42 inhibits myelin sheet formation in vitro (Horiuchi et al., 2012). Aβ_1–42 directly binds myelin basic protein (MBP; Liao et al., 2009, 2010; Kotarba et al., 2013). There is a focal loss of oligodendrocytes and myelin within and adjacent to amyloid plaques in familial and sporadic AD brain (Mitew et al., 2010). However, the absence of MBP decreases the accumulation of Aβ_1–42 in transgenic AD mice, and essentially eliminate amyloid plaques (Ou-Yang and Van Nostrand, 2013). More recently it has been discovered that loss of ceramide synthase 2 activity, necessary for myelin biosynthesis, precedes tau and amyloid pathology in human AD cortex (Couttas et al., 2016). In addition, oligodendrocyte and myelin injury can precede the formation of amyloid plaques and tau pathology in a mouse AD model (Mitew et al., 2010; Hall and Roberson, 2012). Clinically, WMH are more highly associated with preclinical AD than imaging and cognitive markers of neurodegeneration; and WMH are now considered a core feature of dominantly inherited AD (Desai et al., 2010; Kandel et al., 2016; Lee et al., 2016). Of interest, there are very high titers of autoantibodies against a variety of myelin proteins in blood of AD patients compared to controls (Papuc et al., 2015). The cause of the myelin injury in AD brain, however, remains to be elucidated. This review hypothesizes that Gram negative bacterial LPS molecules are present in AD WM and GM where they bind oligodendrocytes and cause an increase in cytokines and oxidative stress that contributes directly to damage of oligodendrocytes and to myelin proteins including MBP which then associate with Aβ_1–42 in amyloid plaques in AD brain.

There is emerging evidence that the gut microbiome affects neurological diseases including AD. There is a different gut microbiome in wild type mice compared to mouse AD models (Shen et al., 2017) and in control compared to AD patients (Vogt et al., 2017). There is an association of brain amyloidosis with pro-inflammatory gut bacterial taxa and peripheral inflammation markers in cognitively impaired elderly humans (Cattaneo et al., 2017). As people age they have more Gram-negative bacteria in the gut. Antibiotic-induced perturbations in gut microbial diversity influences neuro-inflammation and amyloidosis in a murine model of AD (Minter et al., 2016). Serum IgG antibody levels to periodontal microbiota, some derived from the gut, are associated with incident AD (Noble et al., 2014). The gut microbiome is implicated in normal neurodevelopment, autism spectrum disorders, schizophrenia, depression, Parkinson’s, multiple sclerosis, stroke and aging (Branton et al., 2016; Sampson et al., 2016; Sharon et al., 2016; Winek et al., 2016). Though there is a marked increase in reports on the gut microbiome affecting neuropsychiatric diseases, very few have developed a cogent hypothesis on how this occurs. This proposal hypothesizes that Gram-negative bacteria, potentially from the gut or gums as well as from systemic infections, all release LPS which is engulfed via TLR4-CD14 receptors by blood leukocytes including monocytes and neutrophils, and by brain microglia. The TLR4 mediated activation of NFkB increases cytokines which contribute to myelin damage. Studies have demonstrated that NFkB pathway regulates the activity of Beta-secretase 1 (BACE1; Buggia-Prevot et al., 2008; Guglielmotto et al., 2012), the crucial enzyme for Aβ production and the BACE1 level in AD brain is increased compared to control (Guglielmotto et al., 2012). These studies suggest that activation of NFkB may contribute to increases of Aβ and amyloid pathology in AD brain.

The combination of LPS-IS-HY demonstrated evidence of myelin injury in the adult rat brain (Zhan et al., 2015a). Thus, a search for evidence of myelin injury in AD compared to control brains was undertaken, and determined if LPS was associated with the myelin injury in AD brain.

MBP and dMBP levels were examined in 13 AD brains and 10 control brains (Figure 4; Zhan et al., 2015b). dMBP was found in every AD brain, but not in every control aging brains (Figure 4A). Quantification showed more MBP in AD compared to control brains (Figure 4B), and much more dMBP in AD compared to control brains (Figure 4B). Indeed, the ratio of dMBP/MBP was greater in AD compared to control brains (Figure 4B; Zhan et al., 2015b). The antibody to dMBP stains a protein at 37 kDa which is higher in molecular weight compared to intact MBP. It was postulated that the antibody to dMBP probably detects some fragment of MBP (degraded MBP/dMBP) which complexes with other molecule(s) (Zhan et al., 2015b).

The next studies determined whether LPS co-localized with MAG (myelin associated glycoprotein) stained oligodendrocytes in cortex (Figure 5; Zhan et al., 2016). There appeared to be more MAG stained oligodendrocytes in AD (Figure 5B2) compared to control cortex (Figure 5A2). There were more LPS stained cells in AD (Figure 5B1) compared to control (Figure 5A1) cortex. Most LPS stained cells in cortex co-localized with MAG stained oligodendrocytes (Figures 5A3,B3), with more LPS/MAG stained oligodendrocytes in AD (Figure 5B3) compared to control cortex (Figure 5A3; Zhan et al., 2016). These data show more oligodendrocytes in AD brain compared to control and could explain higher MBP levels in AD brain (Figure 4B). In addition, since LPS co-localizes with MAG stained oligodendrocytes, this suggests LPS could damage oligodendrocytes leading to increased dMBP seen in AD brain (Figure 4B). LPS acts on the TLR4 receptor to activate NFkB which increases cytokines that can damage oligodendrocytes and damage myelin proteins (Pang et al., 2003, 2010; Deng et al., 2008, 2014; Paintlia et al., 2008). Not shown here is an increased number of oligodendrocyte progenitor cells (OPCs) in AD compared to control brain (Zhan et al., 2016). The data suggests that LPS injures oligodendrocytes and myelin proteins including dMBP and MAG, and that this leads to proliferation of OPCs that differentiate into mature oligodendrocytes (Zhan et al., 2016).

With evidence of consistent myelin injury in AD brain, the localization of dMBP in AD brain was examined. Neither MBP (Figure 6A2) nor neurofilament protein (NF; Figure 6B2) co-localized with FSB ((E, E)-1-fluoro-2, 5-bis (3-hydroxycarbonyl-4-hydroxy) styrylbenzene, Figures 6A1,B1) stained amyloid plaques in AD brain (Figures 6A3,B3). However, Aβ_1–42 (Figure 6C1) did co-localize with dMBP (Figure 6C2) in amyloid plaques in AD brain (Figure 6C3; Zhan et al., 2015b). These data were supported by biochemical findings suggesting dMBP directly bound AβPP and Aβ_1–42 (Zhan et al., 2015b). Figure 6D shows that immunoprecipitation with an antibody to dMBP followed by Western blotting with an antibody to AβPP showed bands in AD and control brain (Figure 6D), but with greater levels in AD brain (Figure 6E; Zhan et al., 2015b). These data support the idea that LPS injures oligodendrocytes with the resultant dMBP binding to and being co-localized with Aβ within amyloid plaques in AD brain.

The distribution of dMBP in WM of AD brain was also examined (Figure 7; Zhan et al., 2014). Using an antibody to galactocerebroside (GALC) as a marker for myelin, many GALC stained walls of vesicles in periventricular WM (PVWM; Figure 7A1) and the perivascular space (Figure 7B1) of AD brains were found (Zhan et al., 2014). These GALC stained vesicles were shown to co-localize with dMBP in both periventricular WM (Figures 7A2,A3) and in the perivascular space (Figures 7B2,B3) of AD brains. There were fewer vesicles in control brains in the periventricular WM (not shown). The above results show consistent myelin injury in both AD gray and WM (Zhan et al., 2015b).

Gram-negative E. coli bacteria can synthesize extracellular amyloid (Zhao and Lukiw, 2015). This may be relevant since a recent RNAseq study demonstrated bacterial molecules in human brain, and showed the majority were associated with Gram-negative, LPS containing alpha Proteobacteria (Branton et al., 2013). Thus, the LPS found in AD brain could be derived from molecules of Gram-negative bacteria entering brain from the blood, from Gram-negative bacteria entering brain, or possibly from endogenous Gram-negative bacteria in brain.

LPS binds the TLR4/CD14 complex on peripheral monocytes/macrophages or brain microglia to activate NFκB and increase production of cytokines including IL1, IL6 and TNF (Ikeda et al., 1999; Rossol et al., 2011; Enkhbaatar et al., 2015). LPS down regulation of ADAM10 and upregulation of BACE-1/PS-1 may partially explain LPS induced increases β-AβPP and Aβ (Sheng et al., 2003; Zhao et al., 2014; Wu et al., 2015). LPS also increases Aβ levels in brain by acting at the blood brain barrier and impairing LRP which is responsible for Aβ efflux from brain (Erickson et al., 2012; Banks et al., 2015). LPS at high doses can damage the BBB (Banks et al., 2015) which could facilitate entry of LPS itself into brain. Alternatively, monocytes/macrophages may carry LPS into brain. LPS binding to TLR4 on endothelial cells also leads to cytokine release (Verma et al., 2006). LPS binds serum amyloid P and Aβ (de Haas et al., 2000), it binds MBP (Raziuddin and Morrison, 1981) and LPS promotes tau hyper phosphorylation (Kitazawa et al., 2005; Lee et al., 2010; Liu et al., 2016). Thus, LPS could contribute to all of the key neuropathological findings in AD brain including: amyloid plaques, myelin injury and tau hyperphosphorylation (Figure 8).

Polymorphisms in TLR4 and CD14 have been associated with AD in some studies (Balistreri et al., 2008; Rodríguez-Rodríguez et al., 2008; Chen et al., 2012) and whole genome studies have implicated TLR4 in AD (Li et al., 2015). TLR4 levels are increased in AβPP transgenic AD mice and human AD brain and treatment of microglia with amyloid peptide increases IL6 and TNF in microglia and kills microglia via the TLR4 receptor (Walter et al., 2007). The microglial TLR4 receptor is required for recruitment of leukocytes into brain in response to intracranial injections of LPS (Zhou et al., 2006). CD14 binds amyloid peptide fibrils (Fassbender et al., 2004) and is a microglial receptor for phagocytosis of Aβ (Liu et al., 2005). Deletion of CD14 attenuates pathology in AD mice by decreasing inflammation (Reed-Geaghan et al., 2010). In the neonatal LPS/hypoxia model, LPS mediates death of oligodendrocytes via the TLR4 receptor (Lehnardt et al., 2002); and, LPS activated microglia can kill OPCs (Pang et al., 2010).

LPS-TLR4-NFκB mediated increase of cytokines is proposed to damage oligodendrocytes and myelin which leads to formation of myelin aggregates (Figure 8). TLR4 mediated increases of IL1, IL6 and TNF can kill mature oligodendrocytes, oligodendrocyte progenitors and damage myelin (Fan et al., 2009; Xie et al., 2016). Tumor necrosis factor alpha mediates lipopolysaccharide-induced microglial toxicity to developing oligodendrocytes when astrocytes are present (Selmaj and Raine, 1988; Li et al., 2008). Other cytokines are also induced by TLR4 activation and these might also play a role in injury.

MBP directly binds AβPP and Aβ and inhibits Aβ fibrillary assembly via residues 54–64 in MBP (Liao et al., 2009, 2010; Kotarba et al., 2013). Pure MBP degrades AβPP and Aβ peptides, though degraded MBP lacking autolytic activity may not degrade Aβ₄₀ or Aβ₄₂ (Liao et al., 2009). A knockout mouse of MBP (bigenic Tg-5xFAD/MBP−/−) showed markedly decreased numbers of amyloid plaques and decreased insoluble Aβ (Ou-Yang and Van Nostrand, 2013). These findings suggest degraded MBP (dMBP) might play a role in amyloid plaque formation. Axonal/myelin injury results in formation of myelin aggregates, and degraded MBP binds AβPP and Aβ peptides (Liao et al., 2009, 2010; Kotarba et al., 2013), and along with LPS and other molecules leads to the formation of amyloid plaques.

Familial AD is due to mutations in AβPP, presenilin-1 and presenilin-2, and these mutations result in an increase of brain and blood Aβ. It is still debated whether soluble or insoluble Aβ are important, and how they relate to tau hyper phosphorylation. Several groups have proposed that Aβ actions on the TLR4/CD14 complex could be important in the pathogenesis of AD (Figure 8). Aβ binding to TLR4/CD14 would increase cytokines which would lead to further increases of Aβ and to myelin injury and production of dMBP (Figure 8). This could help explain the association of WMH/myelin injury with familial AD (Lee et al., 2016).

To begin to address this, studies were done to determine whether a mouse model of AD which contained 3 AβPP mutations and 2 presenilin-1 mutations (5XFAD mouse) had myelin aggregates associated with its amyloid plaques. Indeed, at 8 months (Figure 9A) and 10 months of age (Figure 9B) the 5XFAD mouse showed that MBP stained myelin aggregates co-localized with FSB stained amyloid plaques (Zhan et al., 2015a). Though it is not known how the myelin is damaged, it is possible that Aβ activation of TLR4 could lead to myelin injury as shown in Figure 8. That is, both LPS and Aβ are ligands for TLR4/CD14 (Erridge, 2010) and thus both could contribute to myelin injury. This conclusion supports recent imaging studies that found WMH to be a core feature of autosomal dominant familial AD (Lee et al., 2016). Thus, sporadic WMH could be a result of the actions of LPS-TLR4 mediated injury and/or Aβ-TLR4 mediated injury to myelin possibly in combination with ischemia/hypoxia.

Though the above discussion focuses on LPS, other Gram-negative molecules are found in AD brain (Zhan et al., 2016). The significance of these to AD pathogenesis is not clear, but could be important. The source of LPS and other bacterial molecules in brain is not clear, so that exogenous infections vs. some source within the body or the brain must be resolved. Since LPS is in AD brain, perhaps molecules from other classes of infectious agents might also be relevant in subgroups of AD cases that might be mediated via other Toll-like receptors.