The Hidden Role of Non-Canonical Amyloid β Isoforms in Alzheimer’s Disease

Recent advances have placed the pro-inflammatory activity of amyloid β (Aβ) on microglia cells as the focus of research on Alzheimer’s Disease (AD). Researchers are confronted with an astonishing spectrum of over 100 different Aβ variants with variable length and chemical modifications. With the exception of Aβ1-42 and Aβ1-40, the biological significance of most peptides for AD is as yet insufficiently understood. We therefore aim to provide a comprehensive overview of the contributions of these neglected Aβ variants to microglia activation. First, the impact of Aβ receptors, signaling cascades, scavenger mechanisms, and genetic variations on the physiological responses towards various Aβ species is described. Furthermore, we discuss the importance of different types of amyloid precursor protein processing for the generation of these Aβ variants in microglia, astrocytes, oligodendrocytes, and neurons, and highlight how alterations in secondary structures and oligomerization affect Aβ neurotoxicity. In sum, the data indicate that gene polymorphisms in Aβ-driven signaling pathways in combination with the production and activity of different Aβ variants might be crucial factors for the initiation and progression of different forms of AD. A deeper assessment of their interplay with glial cells may pave the way towards novel therapeutic strategies for individualized medicine.


Introduction
The neuroinflammatory activity of microglia has, since its discovery, been suspected to contribute to Alzheimer's Disease (AD) [1][2][3]. Early reports referred to an invasion of reactive microglia at the site of senile plaques [4,5]. Moreover, it was observed that both microglia and astrocytes display exacerbated pro-inflammatory activity in disease-affected brain areas [6]. This led to the theory that glial cells are either directly involved in the pathogenesis of AD or are the immune system's defensive shield against it [1]. Recent advances have solidified both presumptions and thus placed the inflammatory activity of glial cells as the focus of AD research [3].
AD is closely linked to the deposition of misfolded amyloid peptides in plaques and the occurrence of neurofibrillary tangles consisting of hyperphosphorylated tau protein [7]. While these discoveries were made more than three decades ago [8], the precise role of Aβ in neuroinflammation and neurodegeneration is still incompletely understood [9]. Progress is hampered by some enormous challenges in Aβ research, ranging from a sophisticated peptide process involving multiple proteases [10,11], to complex physiochemical properties of the resulting peptides [12], leading to different oligomeric structures [13] that ultimately elicit miscellaneous physiological effects on neurons, glia, and immune cells [3,7]. We now argue that these non-canonical Aβ peptides are of importance in the pathogenesis of AD. Firstly, N-abridged species are the most common form of Aβ peptides and comprise about 70% of all Aβ variants found in human brains [8,70]. Second, their levels in the CSF are directly associated with the onset of AD [71]. Third, some chemical modifications such as pyroglutamylated Aβ, which are dominant in AD but not during normal aging, only occur on abridged peptides [72]. Fourth, some familial APP mutations such as the well-known Swedish, French, or German variants alter APP processing and the ratio of abridged Aβ variants, which may directly influence the progression of AD (see Figure 2B) [73]. Table 1. Overview of abridged and/or modified Aβ-species detected in the brain or CSF of AD patients. Superscript description indicates modification at the respective amino acid (glyco-Y10 = glycosylation at tyrosine at position 10 of the Aβ-domain sequence, ox-M35 = oxidation at methionine at position 35, pyro-E3/11 = pyroglutamylation at glutamate at position 5 or 11, race-D-S26 = racemization of D-conformation serine at position 26).

Aβ Fragments
Modification Source References
β-secretase activity leads to the generation of the classical variants Aβ 1-42 and Aβ 1-40 mainly via BACE1 (β-site amyloid precursor protein cleaving enzyme 1). BACE1 produces membrane-bound β-C-terminal fragments (β-CTF) by cleaving APP at position 1 (β-site) or at position 11 (β'-site) of the Aβ-domain that are subsequently further processed either by αor γ-secretases. Cleavage by an α-secretase produces p3 peptides and corresponding Nterminal fragments Aβ 1-14 to Aβ [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16] . γ-secretase activity is mediated by multiple proteins such as APH-1, PEN-2, Nicastrin, and Presenilin-1 or Presenilin-2 [10]. Peptides that result from γ-secretase activity end with residues ranging between position Aβ x-37 to Aβ x-43 . This process is thought to occur through multiple separate intermediate cleavage steps, where mostly three amino acids are cut off within each step [10]. The cleavage process usually starts at position 48 or 49, continues at position 45 or 46, and ends at position 38, 40, or 42. More rarely, peptides ending at position 34, 37, 39, or 43 are generated [10]. The C-terminal variability is due to the imprecise cleavage of γ-secretases whose recognition motifs are not depending on an exact amino acid sequence. In the healthy brain, 50% of the peptides generated this way end at position 40, 16% at position 38, and 10% at position 42 [99]. These ratios shift in AD, leading to a higher production of peptides ending at position 42 [10]. The reason for this is not fully understood; however, mutations of either APP or γ-secretase enzyme complexes can lead to a lowered binding of Aβ to the protease complex, which may then result in a premature release of the cleaved peptides [10]. In addition to these longer variants, γ-secretase can also generate shorter peptides (Aβ 1-14 to Aβ 1-16 ); however, the exact pathways have not been identified, yet [100]. In addition to processing by traditional secretases, Aβ and APP are targeted by a wide range of enzymes and peptidases that can either cut APP directly or interact with Aβ peptides after their generation (see Figure 2B) (reviewed in [11]). These include matrix metalloproteinases such as MMP2 and MMP9, which mainly interact with extracellular or internalized Aβ in glial cells or deposition of fibrillary Aβ aggregates or others such as meprin-β that cleave APP directly in its membrane bound form [11]. This leads to an extensive repertoire of possible peptides (see Table 1).

N-Abridged Aβ Species
Aβ 2-X and Aβ 3-X are thought to be generated by APP cleavage by the metalloprotease meprin-β or after cleavage of "full-length" Aβ 1-X by the exopeptidase aminopeptidase A [101]. Not much is known about the differences between Aβ 2-X and non-abridged variants regarding their biological effects. However, Aβ 2-40 shows a higher aggregation propensity than Aβ 1-40 and can seed further aggregate formation of Aβ 1-40 [102]. Cleavage by meprin-β depends on a specific recognition motif in APP that includes amino acids neighboring the N-terminus of Aβ [102]. This motif is altered in some APP mutants such as the Swedish APP variant, which leads to a diminished processing by these enzymes and a lower number of these variants in patients that carry this mutation [102]. Of note, many popular humanized AD mouse models such as 5xFAD and APP/PS1 mice express the Swedish APP mutation [92]. Thus, these mice have a low amount of Aβ 2-X . By contrast, these N-terminal truncations are frequent in human AD patients [103]. Aβ 4-X variants are abundantly found in dense amyloid plaque cores of human AD patients and 5xFAD mice [104]. Interestingly, Aβ 4-X peptides appear to be more commonly released by microglia and astrocytes than by neurons, but the exact processing mechanism is currently unknown [98]. Several studies suggest that a disintegrin and metalloproteinase with thrombospondin motifs (ADAMTS4) and neprilysin may generate these truncations [104][105][106]. ADAMTS4 has been implicated in the generation of variants such as Aβ 4-40 and Aβ 4-42 [105]. Both show increased aggregation rates and a similar neurotoxicity as non-abridged Aβ [95,107]. Clearance of oligomeric Aβ 4-40 and Aβ 4-42 from the brain is markedly reduced compared to their non-abridged counterparts [108], which may indicate reduced or hampered interactions with glial cells. Of note, injections of Aβ 4-40 or Aβ 4-42 into the brain of mice elicit behavioral deficits and impaired memory, which were comparable with those induced by Aβ 1-42 , while injections with Aβ 4-38 had no effect [95]. Interestingly, transgenic mice expressing human Aβ 4-42 in their hippocampus do not develop plaques but still suffer from hippocampal pyramidal neuron loss and behavioral deficits [95,109].
Aβ 11-X peptides are generated by the β-secretase processing pathway after cleavage at the β'-site. Aβ  and Aβ 11-42 levels are commonly elevated in the brain of AD patients [70,79]. They are approximately as abundant as Aβ 1-42 and can comprise up to 20% of the peptide content in plaques [70,79]. Decreased levels of Aβ 11-X in the CSF are associated with the onset of symptoms in AD and have thus been proposed as a biomarker for AD [71]. In plaque cores, Aβ 11-X co-localizes with Aβ 1-42 and is therefore believed to contribute to early plaque formation [110]. In vitro studies demonstrated that Aβ  leads to increased neurotoxicity exceeding that of the canonical Aβ 1-40 [111]. Furthermore, Aβ 11-40 is capable of rapidly forming fibrils [112] and can therefore act as a cross-seed for the deposition of other Aβ species in plaques [111]. Interestingly, homogenous aggregation of Aβ 11-40 decreases its neurotoxicity while heterogeneous aggregates of Aβ 11-40 and Aβ 1-40 are even more harmful than monomeric or homogeneous oligomers of Aβ 1-40 [111]. Surprisingly little is known about the effects of Aβ 11-40 on glial cells. However, we recently showed that Aβ  activates the human glial cell line U87 via FPR1 [113].
Aβ  and Aβ 17-42 are p3 peptides that are generated via the α-secretase processing pathway. P3 peptides are found in diffuse plaques of AD patients [114] and microglia of AD patients contain various N-terminally abridged Aβ species [115]. Cell culture experiments indicate that they might be produced in even higher amounts than Aβ 1-42 and Aβ 1-40 [116]. Of note, the presence of Aβ 17-X in the CSF is positively correlated with cognitive impairment of AD patients [71], which argues for a contribution of these peptides to AD. However, they are difficult to detect because many commonly used Aβ antibodies are not selective for Aβ 1-42 and cross-react with p3 fragments and other Nterminally abridged variants [117]. In addition, many established protocols for isolating Aβ peptides from brain and CSF samples are not suitable for p3 peptides due to their high hydrophobicity and insolubility, which excluded them from many studies that evaluated Aβ levels [118]. Moreover, since several non-canonical proteases such as IDE, NEP and MMP9 have cleavage motifs within the p3 sequence ( Figure 2B), it is likely that additional p3 peptides with highly variable C-termini exist. However, these potential variants have not been studied so far. Historically, Aβ 17-X were classified as non-amyloidogenic because they did not display fibrillary aggregation and deposition. However, recent advances have cast doubt on these assumptions because conflicting data regarding the amyloidogenic capabilities of p3 peptides exist (reviewed in [94]). While early reports found no evidence of toxic fibril formation by p3 peptides [119], more recent studies found that p3 peptides form fibrils and soluble oligomers [120][121][122] with even faster aggregation kinetics than Aβ  or Aβ 1-42 [107,120]. Aβ  and Aβ 17-42 induce pro-inflammatory cytokine production in human and murine glial cell lines, leading to neuronal decay through pro-inflammatory caspase-activation in neuronal cell lines [107,123], which triggers pro-inflammatory activity after injection into mouse brains [124]. Moreover, we recently showed that all human FPRs can detect Aβ  and that FPR1 is activated at 30-fold lower concentrations by the peptide than by Aβ 1-42 [113].

C-Terminal Variants
Aβ 1-43 is produced by γ-secretase cleavage and can be found abundantly in amyloid plaques [125][126][127]. Its levels are heightened in both spontaneous and familiar AD patients, but not in age-matched controls [128]. In particular, familiar AD cases with mutations in the γ-secretase subunit PS1 show an elevated generation of Aβ 1-43 compared to other species [129,130]. Unlike shorter variants, only low amounts of Aβ 1-43 are found in the CSF [131] and cerebral blood vessels [125]. Aβ 1-43 is highly amyloidogenic and neurotoxic in mouse neuronal primary cultures and cell lines [132,133]. Furthermore, injections of Aβ 1-43 into APP mice lead to severe depositions of Aβ 1-42 , indicating the potent capability of Aβ 1-43 to seed the aggregation of other variants [134].
Aβ 1-37 to Aβ 1-39 are produced by the traditional β-secretase pathway. Compared to canonical Aβ 1-42 and Aβ 1-40 , these C-abridged peptides appear to be harmless or at least significantly less neurotoxic [135,136]. Aβ 1-38 forms oligomers and fibrils, but its low abundance in plaques indicates that it is easily cleared from the CNS. Furthermore, levels of these C-abridged peptides in CSF are altered in patients with AD [137]. Functionally, C-abridged Aβ species can act as a scavenger for Aβ 1-42 and Aβ 1-40 because they can form aggregates with these peptides, leading to reduced synaptic disruption, reduced neurotoxicity, and lowered amyloidogenicity [136]. In a Drosophila melanogaster model, expression of Aβ 1-42 led to severe neuronal and behavioral pathologies, but co-expression of either Aβ 1-37, Aβ 1-38 or Aβ 1-39 attenuated these effects [135]. In addition, a lower ratio of soluble Aβ 1-42 /Aβ 1-38 (i.e., a higher Aβ 1-38 content) was associated with a later age-at-death in male AD patients [136].
Aβ 1-34 is present in the brain and CSF of AD patients and AD mouse models [138,139]. Its levels are highly increased in the CSF of AD patients compared to healthy controls [139]. Interestingly, Aβ 1-34 is also commonly found in brain vessels in early AD stages but diminishes with further progression of the disease [139]. Aβ 1-34 generation processes have not been fully elucidated. In cell culture experiments and transgenic mice, generation of Aβ 1-34 depended on initial BACE1 activity [139], but its C-terminal truncation is likely produced through procession of Aβ 1-40 or Aβ 1-40 by glia-derived metalloproteases such as MMP2 and MMP9 [140]. Thus, Aβ 1-34 is thought to be a degradation marker for clearance activity of microglia in early stages of AD [138][139][140]. Interestingly, an in vitro study reported that, Aβ 1-34 protected APP-expressing HEK293 cells against caspase-3-mediated apoptosis, which may indicate that Aβ 1-34 is not just a side product of degradation processes but may also have resolving properties [141].
Aβ 1-14 to Aβ 1-16 are commonly generated during APP processing but are thought to be harmless [142] because they do not elicit pro-inflammatory glia signaling, are not capable to form amyloidogenic aggregates, and are most likely quickly degraded. Increased levels of these fragments in the CSF indicate increased β-secretase activity and have thus been proposed as a biomarker for Aβ production [78,100].

Post-Translational Aβ Modifications (PTMs)
Pyroglutamylation refers to the conversion of glutamate into pyroglutamate, which can occur in N-abridged Aβ peptides at position 3 (Aβ 3-x pyro-E3 ) and position 11 (Aβ 11-x pyroE11 ). This process was shown to be catalyzed by glutaminyl cyclase in vitro [143,144] and also in vivo [145,146]. Pyroglutamate highly influences the secondary structure of Aβ, leading to an enhanced β-sheet structure, higher hydrophobicity, and increased aggregation propensity [147]. Aβ pyroE species are peculiarly abundant in senile plaques [78,148] and are also present in diffuse, soluble aggregates [149,150]. Aβ 3-X pyro-E3 is highly amyloidogenic and can seed the aggregation of other Aβ species, which increases fibril formation significantly [151,152]. Such cross-seeded aggregates appear to be more neurotoxic than homogenic aggregates and may thus contribute to the pathological effects of AD [152]. Unlike most Aβ variants, Aβ pyroE is only associated with the progression of AD but not with ageing [72]. In pre-clinical AD patients without senile plaque formation, Aβ pyro is not initially present within soluble aggregates [149] but appears during further progression of the disease when patients experience their first cognitive symptoms [149,153]. Deposition of Aβ  pyro-E3 into insoluble plaques precedes deposition of Aβ 1-42 and Aβ 1-40 [148] and may thus be involved during the onset of early symptoms. Aβ 3-X pyro-E3 species also accumulate in lysosomes of microglia [154] where they are suspected to cause lysosomal failure, microglial decay, and subsequent progression of AD [155,156]. In addition, pyroglutamylation may affect protease-mediated decay, and influence the capabilities of Aβ to interact with cell-surface receptors. For example, murine NMDAR recognizes Aβ but not Aβ 3-X pyro-E3 [157]. Furthermore, experiments with porcine primary microglia revealed that Aβ  pyro-E3 is more potent to enhance E. coli phagocytosis than the non-abridged peptide (360). Unfortunately, production and deposition of Aβ pyroE in most AD animal models is not comparable with human AD patients because in APP transgenic mouse lines initial plaques are devoid of Aβ 3-X pyro-E3 species [158][159][160]. Nonetheless, in some cases mouse models can provide valuable insights into the biological effects of Aβ pyroE species. For example, 5xFAD mice overexpressing the human glutaminyl cyclase show an increased Aβ pyroE production, plaque formation, and behavioral symptoms, while the glutaminyl cyclase knockout reduces plaque deposition and rescues behavioral perfor-mance [145]. Moreover, treatment of APP/PS1 mice with antibodies against Aβ 3-X pyro-E3 ameliorates behavioral symptoms and decreases amyloid plaque numbers, which is likely due to FcR-mediated clearance by microglia [161].
Isomerization of Aβ occurs mainly at asparagine (N) and aspartate (D) residues and is commonly detected at position 1 (Aβ iso-D1 ) and 7 (Aβ iso-D7 ) [162]. Isomerization happens through spontaneous, non-enzymatic reactions. Thus, its probability increases throughout the lifetime of Aβ [163]. In plaque core preparations, Aβ species with iso-D1 and iso-D7 are more common than other variants and are present in higher concentrations than in "younger" diffuse plaques or vascular depositions [164,165]. Of note, isomerization has a strong impact on the secondary structure of Aβ and leads to faster aggregation and deposition [166,167]. Aβ iso-D7 also shows an increased neurotoxicity through excessive NO generation [168,169] and inhibits the α7 Nicotinic receptor, which has been implicated in long-term memory formation [168]. Microglia can internalize isomeric Aβ [164]; however, isomeric Aβ is more resistant against degradation in microglial lysozymes [163] and against intracellular and extracellular proteases [163,170], which fosters lysosomal failure [155,156].
Phosphorylation can potentially occur at serine residues 8 (Aβ pho-S8 ) and 26 (Aβ pho-S26 ) and at tyrosine residue 10 (Aβ pho-T10 ) [13]. These modifications are thought to occur mainly through extracellular kinases or after internalization. Aβ pho-S26 accumulates in early AD stages in neurons, but only low concentrations are present in extracellular plaques [171,172]. It can assemble into soluble oligomers that do not form fibrils and exhibit increased neurotoxicity in cell culture experiments [172]. Aβ pho-S8 levels are elevated in later stages of AD where it mainly occurs in compact plaques [173]. A phosphorylated serine at position 8 increases the stability of Aβ pho-S8 aggregates by attenuating their recognition by degrading enzymes and inhibiting microglial clearance [174,175]. Due to the clearance resistance, Aβ pho-S8 species may contribute to plaque spreading [175]. In this context, Hu and colleagues showed that Aβ  pho-S8 can cross-seed with non-modified Aβ 1-42, which generates aggregates with elevated neurotoxicity [152].
Oxidation is mainly caused by radicals such as ROS and NO and can occur at the methionine (M) at position 35 (Aβ ox-M35 ). Aβ ox-M35 molecules are abundantly released by reactive microglia. Oxidation disrupts fibril formation and destabilizes oligomer formation [176,177]. Since generation of oxidative stress is a hallmark of neuroinflammation, the amount of oxidized Aβ is thought to increase during the progression of AD. Oxidized Aβ species with varying length have been identified in brain and CSF samples of AD patients [70,74,80]. A study by Head and colleagues found oxidized species in 46% of diffuse plaques and in 98% of cored plaques [178]. Oxidized Aβ was found in plaque-invading microglia of AD patients [178]. However, so far it has not been elucidated if oxidation influences interactions between Aβ and microglia.
Racemization converts amino acids from their L-into D-conformation. For Aβ, racemization was observed in aspartate residues (D) at position 1 (Aβ D-D1 ) and in seryl residues (S) at position 26 (Aβ D-S26 ) [13]. Racemization of aspartic acid increases the fibrillary aggregation kinetic [179]. In addition, further modifications such as sylation and nitration were observed [13]. However, their biological relevance is yet unclear. Some studies suggest that these PTMs may also influence the aggregation properties of Aβ and could influence cross-seeding with other Aβ species [152].

Splice Variants of APP
The APP gene is highly conserved in mammals, consists of 18 exons, and spans over 290 kilobases [89,180]. Ten APP splice variants have been described that range between 639 to 770 amino acids [89]. The three variants APP695, APP751, and APP770 make up almost the complete protein quantity [181,182]. APP770 is the longest isoform and contains all exons [181,182]. In comparison, APP751 does not include exon 7, which encodes for an additional Kunitz-type protease inhibitor (KPI) domain that protects the protein against proteolytic cleavage by certain enzymes [183]. APP695 lacks both exons 7 and 8,and is therefore deficient of the KPI domain from exon 7 and an OX-2 domain from exon 8, which is thought to influence cell-surface binding [184]. All three mayor isoforms contain the full Aβ sequence and can thus potentially be processed into Aβ and p3 peptides. Interestingly, APP695 is the most common isoform in the brain and is mostly expressed by neurons, whereas APP751 and APP770 are more dominant in glial cells [181,182,184].

APP Processing Is Different in Neurons, Astrocytes, Microglia, and Oligodendrocytes
Neurons are beyond doubt the primary source for the canonical Aβ 1-40 and Aβ 1-42 variants [185,186]. They mostly express the APP isoform APP695 and only minor amounts of APP751 and APP770 [187,188]. Neuronal mRNA levels of APP are approximately 10-fold higher [189] than in other cell types and neurons produce approximately four-fold more soluble APP proteins than astrocytes and microglia [190]. In line with this, Aβ production was also shown to be elevated in neuronal cultures compared to astrocytes or microglia cultures [96]. This is corroborated by a study reporting a~7× higher generation of Aβ 1-40 secreted by neuronal cultures compared to Aβ released from astrocytic cultures [98]. Neurons predominantly secrete Aβ peptides starting at position 1 (80%), whereas those released from astrocytes and microglia represent mainly N-terminally abridged Aβ peptides including Aβ 2/3-x and Aβ 4/5-x ( Figure 3B) [98], suggesting that secretase levels are different in the respective cell types. In line with this, BACE1 protein levels were reported to be more abundant in cultured neurons compared to astrocytes, whereas BACE2 showed a higher presence in astrocytes [191]. Furthermore, neurons produce 3.4 times more Aβ 17-x than Aβ 1-40-43 , while astrocytes generate 7.8 times more Aβ 17-x than Aβ 1-40-43 , indicating a higher rate of α-secretase cleavage in astrocytes compared to neurons [96]. Protein expression of PS1, the catalytic domain of the γ-secretase complex, was shown to be comparable in neurons and astrocytes, suggesting no major difference regarding γ-secretase processing in both cell types [191].
In astrocytes, all three major splice forms of APP, APP695, APP751, and APP770, were shown on the protein level [192]. Splice forms APP751 and APP770 seem to be predominant [190,192], while APP695, which is the major described APP splice form in neurons [187,188], has a lower abundance. This might contribute to differences regarding the proteolytic conversion of APP between astrocytes and neurons. In contrast to neurons, only 40% of the secreted Aβ peptides are cleaved at position 1 in astrocytes [98]. This is of interest, because N-terminally abridged Aβ species are prevalent in neuritic plaques [77,95,193,194]. N-terminally abridged Aβ species from astrocytes and microglia may therefore contribute to a higher proportion to the formation of neuritic plaques than peptides derived from neurons. In line with this idea, cell culture experiments demonstrated that astrocytes and microglia indeed secrete higher amounts of N-terminally Aβ variants such as Aβ 2/3-40 and Aβ 4/5-40 [98]. Proteolytic conversion of these variants does not depend on BACE1 activity but is likely mediated by plasma-membrane associated cathepsin B (CatB) [195]. However, a number of other proteases such as meprin β [196], neprilysin [197], myelin basic protein [198], the metalloproteinase ADAMTS4 [105], and aminopeptidases [199] may also contribute to the production of these N-terminally modified Aβ variants [11,200]. By contrast, the proportions of the C-terminal Aβ variants Aβ 1-37 , Aβ 1-38 , Aβ 1-39 , and Aβ 1-42 to Aβ 1-40 did not differ between neurons, astrocytes, and microglia [98], indicating no differences in the C-terminal Aβ peptide-trimming by γ-secretase between these different cell types [10]. Cortical astrocyte cultures have lower BACE1 protein levels and higher levels of BACE2 than corresponding neuron cultures [191]. In line with the finding that APP is processed by BACE2 at position 19-20 and 21-22, Western blot experiments demonstrated a more efficient generation of Aβ 1-15 , Aβ 1-19 , and Aβ 1-20 in astrocytes of APP/PS1 mice [191]. However, the isolation of primary astrocytes typically yields cultures with mostly reactive phenotypes [201], and these therefore possess presumably higher APP protein levels than resting astrocytes since it has been shown that APP expression in reactive astrocytes is substantially increased following neuronal damage [202]. By contrast, only little APP expression is documented in resting astrocytes in vivo [203]. Therefore, primary cultures may not exactly reflect the in vivo situation of the cell types analyzed. Of note, BACE1 protein and APP levels can be induced by 300 to 600% through stimulation of astrocytes by cytokine combinations or Aβ 1-42 [204].
In microglia, APP protein expression was reported at an early stage [205,206]. All three major splice forms, APP695, APP751, and APP770, are present in microglia cell cultures [192]. However, microglia express more than 50% of their APP mRNA as transcripts as APP695, which encoded for the KPI-domain; approximately 22% of total APP mRNA represented APP770 mRNA, 45% APP751/L-APP752 mRNA, 25% L-APP733 mRNA, and approximately 4% APP695 and L-APP677 [207]. Furthermore, BACE1 expression has been indicated on a protein as well as the mRNA level in microglia cells [208]. The presence of α-secretase ADAM10 in microglia was also shown via an upregulation of ADAM10 after reduction in cortical activity [209]. Interestingly, protein amounts of gamma secretase subunits such as PS1 and Nicastrin are increased in activated microglia [210].
The whole APP processing machinery is present in microglia under certain conditions. However, the impact of microglia on APP processing has not been characterized extensively, yet. Microglia produce mainly N-terminally modified Aβ peptides including Aβ 2/3-x and Aβ 4/5-x [98], suggesting a possible contribution of N-abridged Aβ species from microglia to the peptides found in vascular and neuritic plaques [77,194]. Several studies with the immortalized mouse microglial cell line BV-2 revealed an interaction of microglia with the extracellular matrix that affects APP secretion as well as the intracellular biogenesis of APP [207]. APP expression and release of soluble APP was highest after adherence to uncoated plastic surfaces [207]. A further study using this cell line showed that Aβ [25][26][27][28][29][30][31][32][33][34][35] and lipopolysaccharide treatment induced Aβ secretion [211]. APP is expressed on the surface of microglia cells and it is upregulated due to their activation [205,206]. Direct stimulation of APP with agonist antibodies also led to robust activation of microglia [212,213]. Even an activation of microglia via Aβ-binding to APP has been indicated [214]. Collectively, these data indicate that APP itself can serve as a cell-surface receptor that helps regulating microglial activation.
In oligodendrocytes, analysis of mRNA transcripts identified the APP splice forms APP695, APP751, and APP770 [215]. Interestingly, a lower molecular weight form of full-length APP was more substantially expressed than higher molecular weight forms, suggesting that oligodendrocytes, similar to neurons [216], may express more APP695 than APP751 and APP770 [217]. Adult oligodendrocyte progenitor cells (aOPCs), which were purified from a >4-month-old rat brain, generally expressed BACE1, APP, and components of the γ-secretase complex such as Presenilin-1, anterior pharynx-defective 1 (APH1), and Nicastrin [218]. APP protein expression in oligodendrocytes has also been shown via immunostaining of rodent brain [219][220][221], as well as in cell culture [215,217,222]. Moreover, ADAM10 expression has been documented in oligodendrocytes of the developing brain at later embryonic stages [223]. APP processing analyses in a cell line of murine OPCs (Oli-neu) [224] documented generation of the APP intracellular domain (AICD) by γ-secretase and production of an APP C-terminal fragment, which increases after treatment with a γ-secretase inhibitor [225]. A more detailed study even addressed various differentiation stages of cultured aOPCs and APP processing [218]. These aOPCs can be maintained for several months in media containing Fibroblast growth factor 2 (FGF2) which promotes survival and proliferation of aOPCs positive for NG2 (chondroitin sulfate proteoglycyan NG2/CSPG4), a marker for aOPCs [218]. Interestingly, FGF2 withdrawal increased expression of full-length APP as well as Aβ  production. aOPC cultures including only 0 or 1 ng/mL FGF2 secreted~four-fold higher ratios of Aβ x-42 to total Aβ than cultured fetal rat neurons, suggesting a high rate of γ-secretase processing in these aOPC cultures [218]. α-CTF accumulated in media without FGF2 while β-CTF and the p3 fragment increased with elevated FGF2 concentrations, suggesting that γ-secretase prefers the APP α-CTF as a substrate at higher concentrations of FGF2. Further proteases that are capable of processing Aβ have also been found in oligodendrocytes. One example is ADAMTS4, which is exclusively expressed in oligodendrocytes in mouse brain and generates Aβ 4-x from both full-length APP and Aβ 1-x [105]. Mass spectrometric profiles of different Aβ species from 5-day-old OPC cultures from WT and ADAMTS4 KO mice confirmed release of Aβ 4-x species from these cells [105]. Interestingly, ADAMTS4 mRNA levels were increasing during the culture period of 5 days, suggesting a higher conversion of APP by ADAMTS4 in myelinating oligodendrocytes [105]. Oligodendrocytes therefore seem to be a source of secreted Aβ 4-x peptides [105], which corresponds to the finding of N-terminally abridged Aβ variants such as Aβ 2/3-40 and Aβ 4/5-40 secreted by microglia and astrocytes in contrast to neurons [98].

Contribution of Aβ Internalization and Degradation
Microglia help to remove toxic peptide Aβ isoforms from the CNS via internalization and subsequent degradation by several mechanisms (see Figure 4) [226]. Various microglial receptors are able to mediate uptake of Aβ by endocytosis via formation of receptor-peptide complexes. However, different aggregates seem to preferentially interact with different types of cell-surface receptors. For example, FPRs help in the clearance of Aβ 1-42 monomers and small oligomers [227][228][229], whereas fibrillary aggregates are not efficiently internalized [230,231]. Scavenger receptors such as SR-A [232], SR-B1 [233], and CD36 [234,235] appear to prefer the uptake of fibrillary forms of Aβ 1-42 , whereas TLRs have been reported to mediate the internalization of monomeric, oligomeric, and fibrillary Aβ 1-42 [226]. TLR2 [236,237] and TLR4 [238] induce the uptake of Aβ 1-42 and trigger internal degradation processes. In addition, microglia can also internalize Aβ 1-42 via unspecific uptake mechanisms, such as pinocytosis through phosphoinositid-3-kinase (PI3K) [239] or Ras-related C3 botulinum toxin substrate 1 (Rac1)-dependent pathways [240]. Internalization leads to different internal degradation processes in microglia. In cell culture experiments, fibrillar Aβ 1-42 was effectively degraded through autophagy by lysosomes [241][242][243]. The acidic environment in lysosomes enables the formation of compact Aβ 1-42 aggregates that are resistant against further degradation [155,156]. Microglia can release these aggregates as microvesicles into the extracellular space where they contribute to plaque formation and neurotoxic effects [155,156]. This can trigger a vicious cycle of apoptotic pathways [155] involving an aberrant activation of the NLRP3 inflammasome [244] and the apoptosis of microglia [155]. Some studies suggest that the release of degradation-resistant Aβ 1-42 from dying microglia might be the driving force behind plaque deposition since microglia-deficient animal models show only minimal Aβ 1-42 plaque formation [245,246]. Interestingly, autophagy of Aβ in microglia seems to become less efficient with disease progression [247].

Contribution of NLRP3 Signaling
The wide range of interaction partners for Aβ leads to a number of equally diverse and complex interactions between different signaling cascades in the inflammatory response of microglia (see Figures 4 and 5) [3,23]. These cascades are mainly regulated by inflammasomes, which are cytosolic protein complexes that mediate inflammatory responses of immune cells [249]. In AD, formation of the NLRP3 inflammasome and subsequent caspase-1 activity have been proposed to be driving forces in the microglial response to Aβ [248]. The main components of this inflammasome are the cytosolic receptor NLRP3, the adaptor protein adaptor protein apoptosis-associated speck-like protein containing a caspase activation and recruitment domain (ASC), and the inactive precursor form of caspase-1 (pro-caspase-1) [248,250]. Priming and assembly of the NLRP3 inflammasome is initiated by an activation of NF-κB pathways that are typically induced by extracellular stimulation via surface receptors [248] such as TLRs [236], CD36 [251], and FPRs [252]. This leads to NLRP3 upregulation, which then recruits ASC and pro-caspase-1 for inflammasome formation [3,248]. Upon a second signal, which is either triggered by cell-surface receptors or through direct interaction of NLRP3 with internalized Aβ [253], the primed inflammasome complex autocatalyzes pro-caspase-1 through the adaptor protein ASC into its active state, which subsequently triggers transformation of interleukines and cytokines into their biologically active forms [248]. Of note, microglia release inflammasome components such as ASC specks into the extracellular space, where they can be a seed for Aβ, which may propagate plaque formation [254]. In general, detection of Aβ induces the release of Il1β, IL6, IL8, IL33, IL34, and other inflammatory factors such as TNFα. IL1 and its isoform Il1β belong to the most prominent representatives because upregulation of Il1β already occurs at the earliest stages of plaque evolution [255]. Il1β release increases APP production in neurons [256,257] and influences APP processing via upregulation of proteases [258][259][260]. High IL1 levels were found in brain tissue [261,262] and cerebrospinal fluid [261] of AD patients and APP/PS1 mice [262]. Il1β induces further self-production in microglia and astrocytes, leading to a self-potentiating inflammatory feedback loop [263]. This is associated with neuronal toxicity, exacerbated synaptic loss and aberrant glial activation.  [3,23]. These processes are in part regulated by the intracellular NLRP3 inflammasome. The release of inflammasome components (ASC specks) may lead to seeding of Aβ and may thus contribute to plaque formation [248].

Contribution of Oxidative Stress
Generation of ROS and radical nitrogen species (RNS) is a typical pro-inflammatory action of microglia to remove extracellular pathogens [3,264]. Microglia can release an oxidative burst, which is usually driven by the transmembrane enzyme complex NADPH oxidase (NOX) [264]. NOX activity is triggered as a secondary effect after the phagocytosis of pathogens or the detection of PAMPS or DAMPS by cell-surface receptors such as TREM2 [265] and FPRs [266]. This leads to the generation and release of ROS into the extracellular space. Unfortunately, this does not only target harmful pathogens but also harms neighboring cells, beneficial proteins, and lipids [264]. The healthy brain combats oxidative stress through a complex antioxidant system, which includes radical scavengers such as glutathione that protect cell organelles from damage and dysfunction [264,267]. However, aberrant release of ROS and RNS can overwhelm this antioxidant defense system, leading to severe damage through lipid peroxidation, protein oxidation, and DNA damage [264]. Figure 5. Aβ-mediated activation of microglia. Direct detection of Aβ (e.g., by TREM2, TLRs, FPRs, RAGE) or indirect stimulation through Aβ-mediated processes (e.g., TRPM2 activation by ROS, detection of autoantibodies against Aβ by FcRs) leads microglia to adopt their reactive phenotype in which they utilize intracellular signaling pathways and metabolic processes to initiate their proinflammatory response.
Generation of oxidative stress has been described as a hallmark feature of AD [7]. Signs of oxidative damage are commonly found in the brain, CSF, blood, and urine of patients [268][269][270][271]. In particular, brain areas afflicted with large plaque and NFT depositions show high levels of oxidated proteins and lipid peroxidation. Next, regulatory factors of NOX are upregulated in AD brain tissue [272,273]. Aβ triggers ROS release through activation of microglial cell-surface receptors. The ROS effects are initially dampened by the antioxidant system, but chronic exposure of cells to Aβ leads to its depletion and thus the subsequent exacerbation of oxidative damage [274]. Aβ-mediated oxidative stress then disrupts synaptic signaling, induces astrocyte activation, and leads with increasing levels to neuronal death. In addition, ROS activates TRPM2, which propagates the proinflammatory activity of microglia and initiates apoptotic pathways in neurons [275]. Apoptotic cells subsequently release additional DAMPs that are perceived by microglia and further stimulate the pro-inflammatory machinery, thus leading to the self-renewal of oxidative stress. Moreover, oxidative stress leads to increased production of Aβ through modulation of APP procession pathways [276][277][278][279]. Treatment of neuronal cell lines with H 2 O 2 increases expression and activity of βand γ-secretases that mediate the production of pathogenic Aβ species [276,278,279]. In particular, γ-secretase activity seems to be susceptible to the effects of oxidative stress and may result in increased β-secretase levels through yet unknown pathways [279,280]. In addition, oxidative stress can directly facilitate the generation of modified Aβ species through oxidation at position 35 or nitrosylation at position 10 [13].

Interaction of Aβ with Other Neuropathological Proteins
Amylin (also known as islet amyloid polypeptide (IAPP)) is a peptide hormone generated by islet β cells in the pancreas that is co-secreted with insulin into the blood stream during regulation of blood glucose levels [281]. Amylin-Aβ aggregates are present in cerebral plaques of patients with familiar AD [282]. In CSF samples of familiar AD patients, high Amylin concentrations are associated with decreased Aβ 1-42 levels, which suggests that Amylin may influence Aβ 1-42 transport between the brain, CSF, and blood [282,283]. Interestingly, the Amylin receptor may also be able to interact with Aβ [284]. Amylin and Aβ 1-42 share a sequence homology of more than 50% and form similar β-sheet structures. Moreover, Amylin can act as a seed for Aβ  , which can lead to amorphous heterocomplexes that display a highly increased neurotoxicity compared to either pure Amylin and Aβ 1-42 aggregates [285].
Prion protein (PrP) is a cell-surface protein encoded by the PRNP gene, which is ubiquitously expressed throughout human tissues, but is most abundant in the brain (reviewed in [286]). Here, PrP is mainly expressed in astrocytes and neurons, but low levels are also present in microglia and oligodendrocytes [287][288][289]. In addition to its membraneanchored isoform, soluble PrP variants have been described [290,291]. PrP's biological functions are not fully elucidated but several studies argue for PrP contribution to synapse formation and neuronal signal transduction [292][293][294]. Its physiological form is dominated by an α-helical secondary structure and is referred to as PrP c (cellular form) [286]. However, under pathological condition, PrP undergoes a structural shift into a neurotoxic isoform with predominant β-sheet conformation, also called PrP sc (scrapie form) [295]. Formation of PrP sc is thought to be contagious, leading to further conversion of PrP c into their pathogenic form [296]. PrP sc form aggregates with itself and other proteins that are highly neurotoxic and lead to severe disruption of normal CNS functions [295]. PrP is commonly found in the diffuse and dense plaques of AD patients and where it can be co-localized with Aβ [297][298][299]. In addition, PrP can directly bind Aβ in both its soluble and membranebound form [300,301]. However, the biological consequences of these interactions are still debated because deletion of the PRNP gene did not ameliorate plaque formation in APP/PS1 mice but ameliorated behavioral and cognitive symptoms [302,303]. Interestingly, PrP was shown to modulate BACE1 activity and thus the production of Aβ, but had only diminished effects on processing of the Swedish APP variant [304,305].

Genetic Variants Affecting Microglial Response to Aβ
Microglia and other glial cells express various receptors and molecules that can either directly bind Aβ at the cell surface, which leads to altered cell signaling, or interact with it indirectly, which modifies Aβ detection or degradation ( Figure 5).
TREM2 is a cell-surface receptor that is highly expressed in microglia. It mostly detects phospholipids and glycolipids that are released from host-derived apoptotic cells or from invading pathogens (Yeh 2017), but is also able to interact with multiple peptides and proteins such as Aβ and various Apolipoproteins (Apo) including APOE and APOA1 [306]. TREM2 is capable of inducing cell activation and migration towards Aβ 1-42 [307]. In TREM2-deficient AD mouse models such as modified APP/PS1 and 5xFAD, microglia do not congregate around senile plaques and thus do not respond to the progressing plaque load [306]. In recent years, this receptor has received much attention, since loss of TREM2 leads to nearly complete absence of microglial reactivity in mice [308,309]. In addition, several genome-wide association studies (GWASs) in humans found mutations and polymorphisms of the TREM2 gene to be a high-risk factor for familiar forms of AD (see Table 2).    [354] increase ROS production and Aβ phagocytosis No association with AD described TLRs comprise a family of pattern recognition receptors (PRRs) that detect pathogenassociated molecular patterns (PAMPs) from exogenous sources such as bacteria and fungi, but can also recognize damage-associated molecular patterns (DAMPs) from endogenous sources [355]. There are ten receptor subtypes (TLR1-10) in humans that are all expressed in the central nervous system (CNS) as well as in other tissues [355,356]. In microglia, TLRs induce cell activation after detection of harmful stimuli and initiate the inflammatory cascade which includes the upregulation of additional PRRs and the production of cytokines and reactive oxygen species (ROS) [357]. In the context of AD, TLR2 and TLR4 are of special interest since they are capable to detect Aβ 1-42 . Both TLR2 and TLR4 mediate the phagocytic uptake and degradation of Aβ in microglia [236,358] and induce typical pro-inflammatory cascades such as the release of Il1β and TNFα as a response to the peptides [359]. Their expression is upregulated in inflamed brain tissue and in plaque-invading microglia of human AD patients [359] but also in microglia derived from APP/PS1 mice [238]. APP/PS1 mice with dysfunctional TLR4 display decreased cytokine levels [360] but also more severe plaque formation [238]. Similarly, TLR2-deficient APP/PS1 animals show accelerated cognitive impairment and increased Aβ 1-42 concentrations in the brain [237]. On the contrary, in a study with APP/PS1 mice, long-term administration of TLR2-inhibiting antibodies decreased microglial activity, improved cognitive performance, and lowered Aβ plaque load [361].
CD36 is a class B scavenger receptor that is expressed in microglia, astrocytes, and neurons in the CNS, and in various peripheral cell types such as innate immune cells, myocytes, and endothelial cells [362]. CD36 is mostly known as a fatty acid transporter but is also involved in lipid metabolism and the regulation of inflammatory responses in immune cells [363]. Its ligands include various structural proteins of the extracellular matrix such as collagens and thrombospondins. In addition, CD36 interacts with fibrillary Aβ , which induces a pro-inflammatory activation of microglia [235,364]. Upon detection of Aβ, CD36 mediates cell migration and subsequent binding to Aβ fibrils [364]. It contributes to cytokine production [251] and generation of oxidative stress [234]. In addition, CD36 may act as a co-receptor for other PRRs that interact with Aβ. For example, activation of CD36 can lead to the formation of complexes with TLR2 and TLR6, thus eliciting internalization and degradation of Aβ 1-42 [365]. Of note, in human brain samples, expression of CD36 is detected in microglia proximal to Aβ plaques [366]. However, Ricciarelli and colleagues reported that CD36 is also highly expressed in healthy people with senile plaques [366].
RAGE is a receptor mainly binding advanced glycation endproducts, which are generated by non-enzymatic glycation and oxidation of proteins and lipids [367,368]. Inside the CNS, the receptor is mainly expressed in neurons and glial cells, but can also be found in many other tissues outside the brain [369,370]. Six isoforms of RAGE have been identified. The most common variant is integrated into the membrane (mRAGE), while the other five soluble isoforms lack the transmembrane region (sRAGE) and are thus either cytosolic or secreted into the extracellular space [371]. Moreover, RAGE can be cleaved by ADAM10 and ADAM17, which are also involved in APP processing [18]. RAGE was first associated with AD due to its ability to bind oligomeric Aβ 1-42 , which triggers pro-inflammatory responses in microglia [46,372]. RAGE is involved in Aβ trafficking and clearance through the blood-brain barrier [373] and is thought to modulate secretase activity in neurons, thereby influencing APP processing and Aβ production [374]. Furthermore, RAGE may act as a co-receptor for FPRs that enhances the uptake of Aβ 1-42 in microglia [46]. More recently, a possible contribution to AD through interaction with Amylin (also known as islet amyloid polypeptide) was proposed [375]. In microglia of APP/PS1, mice expression of human RAGE exacerbates the production of pro-inflammatory cytokines such as Il1β and TNFα, increases the production of Aβ 1-42 and Aβ 1-40 , and promotes cognitive decline [376]. In contrast, RAGE-deficient APP/PS1 mice showed a decreased concentration of Aβ  and Aβ 1-40 and improved spatial memory [374].
FPRs are a small family of G-protein-coupled receptors comprising three receptor subtypes (FPR1, FPR2, and FPR3) in humans [377,378] that are expressed in the innate immune system and a number other cell types, including microglia [379]. FPRs mainly detect formyl methionine-containing peptides that occur in bacteria as PAMPs released during bacterial infections and in mitochondria [380,381], where they are released as DAMPs by damaged cells [381]. They also bind some neuropathological peptides such as Aβ 1-42 [231,266] and prion protein fragment PrP 106-126 [382]. In particular, FPR2 has been implicated in AD through its interactions with Aβ 1-42 . In microglia, FPR2 mediates uptake of Aβ 1-42 [227,229,383] and triggers the inflammatory response against it. In inflamed brain areas of AD patients, FPR2 is upregulated in reactive glia [231]. Interestingly, microglia from APP/PS1 mice show an increase in the murine FPR1 and FPR2 [46], and treatment of APP/PS1 mice with a competitive FPR inhibitor ameliorated cognitive impairment, reduced plaque load, and lessened microglial reactivity [384]. The role of the FPR subtypes in the response to non-canonical Aβ species is not well examined. However, our recent study provided the first in vitro evidence that FPRs can detect N-terminally abridged Aβ species, and that FPR1 and FPR3 may also be involved in detection of Aβ 1-42 [113].
MARCO is a Class A scavenger receptor that is mainly expressed in macrophages and glial cells. It mainly binds poly-anionic ligands such as low-density lipoproteins and environmental particles such as TiO 2 and Fe 2 O 3 [385], but has also been shown to capture bacterial PAMPS such as lipopolysaccharides and oxidized lipoproteins, and even helps to engulf whole bacteria in macrophages [386][387][388]. Several studies suggest that MARCO may also bind both fibrillary and non-fibrillary Aβ, either directly or through interactions with other receptors [322]. MARCO modulates intracellular activation of NLRP3 and limits extracellular detection by in glial cells [389]. Furthermore, MARCO can form complexes with FPR2 that facilitate uptake of Aβ 1-42 and may influence FPR signaling [319].
Chemokine-like receptor 1 (CMKLR1) is a G-protein-coupled receptor that is mainly expressed in white adipose tissue and immune cells such as microglia, macrophages, and dendritic cells [390]. CMKLR1 interacts with endogenous ligands such as the adipokine Chemerin, and helps to modulate metabolic processes and cell proliferation, especially during glucose processing, adipogenesis, and angiogenesis [390]. In immune cells, CMKLR1 induces chemotaxis and pro-inflammatory cascades, but is also involved in anti-inflammatory signaling through interactions with pro-resolving ligands such as Resolvin E1 [391]. In brain tissue of AD patients, CMKLR1 is upregulated and co-localizes with Aβ 1-42 [318]. Furthermore, Aβ 1-42 binds to the receptor in an in vitro expression system [318]. In primary mouse microglia and cell lines, interactions between CMKLR1 and Aβ 1-42 induce cell migration, internalization of Aβ 1-42 , and MAPK-dependent inflammatory activation [318]. Knockout of CMKLR1 in APP/PS1 mice leads to increased Aβ deposition but also mortality and cognitive impairment [392]. CMKLR1-deficient mice and wild type mice that were treated with a CMKLR1 inhibitor were more resistant against a chemically induction of Tau hyperphosphorylation [392].
Nucleolin is a phosphoprotein mainly distributed in the nucleolus of many cell types, and is also expressed at the cell surface of macrophages and microglia [330,393]. It can bind DNA, RNA, and amyloid-like proteins [393,394]. Inside the nucleus it has been implicated in regulating DNA and RNA metabolism, chromatin structure, and ribosome assembly [394]. At the cell surface it has been implicated in the cellular-entry of various viruses such as human immunodeficiency virus (HIV) and respiratory syncytial virus (RSV). In primary rat microglia, Nucleolin also robustly recognizes monomeric and fibrillary Aβ 1-42 and mediates its phagocytosis, but shows only weak binding to Aβ 1-40 [330]. Furthermore, Nucleolin interacts with APP mRNA [395], which may result in modulation and increase APP and Aβ production [396,397].
TRPM2 is a non-selective calcium-permeable cation channel [398]. Activation of TRPM2 is thought to contribute to warmth-sensing in neurons [399] and to the regulation of cytokine secretion in immune cells [400]. The ion channel does not directly interact with classical ligands but instead senses mediators of oxidative stress such as ROS and excessive nitric oxide [401]. In microglia, detection of Aβ 1-42 by other receptors leads to the induction of oxidative stress, which in turn activates TRPM2 and induces rapid calcium influx, which triggers cytokine production through the NLRP3 inflammasome [402]. This may lead to a self-renewal of TRPM2 activation since cytokine release can trigger further production of ROS, thus propagating a vicious cycle of pro-inflammatory cascades. Knockout or pharmacological inhibition of TRPM2 in primary mouse microglia attenuates microglial activation by Aβ 1-42 and inhibits production of TNFα [403]. Next, TRPM2deficient APP/PS1 mice show improved spatial memory and reduced microglial reactivity, but no change in plaque load [404].
CD33 is a cell-surface receptor in myeloid cells that is exclusively expressed in microglia within the CNS [405,406]. Through interactions with sialic acids, CD33 can bind glycans and glycolipids. Of note, due to its short extracellular domain, it is assumed that CD33 mainly interacts with ligands on the surface of the cell [406]. In microglia, CD33 does not bind Aβ 1-42 but is thought to interact with other receptors-especially with TREM2-to modulate their signaling and inflammatory responses [407]. Several GWASs have identified CD33 as a high-risk factor for the development of AD (see Table 2). Expression of CD33 is increased in microglia of AD patients and is positively correlated with insoluble Aβ 1-42 levels and plaque burden [408]. Deletion of the CD33 gene in microglia derived from human-induced pluripotent stem cells or in primary mouse microglia improves uptake of Aβ 1-42 [407,408] and leads to increased oxidative burst and production of pro-inflammatory factors [407]. In accordance with these in vitro data, CD33-deficient APP/PS1 mice show reduced brain levels of insoluble Aβ 1-42 as well as amyloid plaque burden [408]. Injection of a viral system encoding microRNA against CD33 in APP/PS1 mice resulted in reduced expression of CD33 and decreased cerebral levels of soluble Aβ 1-42 and Aβ 1-40 [409].
The family of FcRs comprises several receptors that are expressed in microglia and many other immune cells, and can bind the Fc-region of different antibody subtypes. Interestingly, plaque-invading microglia show a strong expression of FcRs [410]. In several animal studies, immunization against Aβ 1-42 led to a significant reduction in plaque load [411,412] and removal of Aβ 1-42 from CSF [413,414]. A possible explanation is that autoantibodies against Aβ 1-42 bind aggregates in senile plaques that are then degraded through FcR-mediated clearance by microglia. However, other studies suggest that these effects are independent from FcRs [415]. During the progression of AD, the bloodbrain barrier is weakened, which may allow peripheral cells and molecules to enter the brain [411,412]. This permits the entry of antibodies that can bind to Aβ aggregates. A subsequent binding of this complex to FcRs on microglia may help to clear these pathological peptides from affected brain areas [413,414] Genome-wide association studies (GWASs) and genetic linkage studies associated with late onset AD (reviewed, e.g., in [416,417]) revealed the importance of genes with a regulatory role for microglia and immune responses. In particular, receptors that are expressed on microglia, such as TREM2 and CD33, have been identified by these approaches. This underpins the key role of microglia in perception of Aβ, the regulation of the brain's inflammatory response to these deleterious peptides, and the removal of the resulting aggregates. For many other putative microglial Aβ receptors (for a summary see Table 2), such as MARCO, identification of AD-related polymorphisms has not yet been carried out.
With regard to "binding to Aβ" of these putative receptors, it has to be taken into account that, in most cases, the direct physical binding has not been proven yet. In most cases, binding has been assumed due to pharmacological or genetic interventions in cells or animals. Exemptions are CD14 or TREM2, where immunoprecipitations were conducted [307,340].
Taken together, these observations argue that a delicate balance of differently modified or processed Aβ variants may finally orchestrate immune responses via a number of different microglial receptors that show a differential presence in different microglia subtypes. A detailed analysis regarding Aβ variant-reactivity or potential selectivity of most receptors is missing to date. Moreover, knowledge on the effect of identified polymorphisms of the receptors regarding recognition of modified/abridged Aβ peptides is even scarcer. Another layer of complexity is added by the fact that some mutations may alter the proteolytic processing, either indirectly through microglial receptor signaling or directly through SNPs in the processing proteases that may affect the cleavage pattern, processivity, and shedding. TREM2 SNPs are a good example of such complex mutual effects. They have been shown to accumulate in cell protrusions in close proximity to Aβ plaques [346] that can be cleaved by the metalloproteases ADAM10/17 C-terminal to histidine 157 [17,[418][419][420]. The amount of soluble (s) TREM2 is increased in the CSF of AD patients and seems to sustain microglial viability, and to trigger inflammatory responses by the Akt-GSK3β-β-catenin and NF-κB pathway in vitro [421,422]. Consequently, its beneficial capacity lowers ApoE4-related risk of cognitive decline [423]. Carriers of the R47H TREM2 polymorphism show higher levels of sTREM2 in CSF than non-carriers [347]. The p.H157Y variant that was identified in the Han Chinese population within the stalk region of the receptor enhances shedding of TREM2 [419,424]. Other variants, such as the TREM2 p.T66M mutation, led to reduced secretion of a soluble partition of the receptor [347,350]. While the amount of R47H TREM2 on the cell surface remained stable despite elevated secretion of sTREM2 [350], uncleaved plasma-membrane tethered p.H157Y TREM2 was found to be severely reduced [419]. An association of the two SNPs within the human α-secretase ADAM10 promoter, rs514049A/C and rs653765C/T, with AD could not be clearly demonstrated [425,426]. However, the latter was found to be correlated with lowered ADAM10 mRNA levels in PBMCs of CC versus CT/TT carriers and with a lowered level of soluble (s) RAGE in plasma [426]. RAGE is also expressed on microglia [427] and processed by ADAM10 [428]; therefore, the ADAM10 promoter polymorphisms may also play a role in related inflammatory processes in the brain, even if this has not been investigated yet. Moreover, the RAGE polymorphism G82S itself influences shedding as a significant association between G82S genotypes and sRAGE plasma concentrations in samples from non-diabetic/non-obese Koreans [429].
Membrane topology and architecture of the putative or identified Aβ receptors on microglia are quite diverse (see Figure 1), and the intense phagocytic activity of these cells, in general, requires a dynamic membrane architecture. Therefore, polymorphisms in the receptor-encoding genes themselves or in genes of the interaction partners may not only orchestrate Aβ perception. Lipidomic analyses of microglial CD11b-positive small extracellular vesicles from the cryopreserved parietal cortex of a restricted number of patients and controls indicated not only increased levels of TREM2, but also a proinflammatory lipid profile in AD (e.g., increase in the most abundant monohexosylceramide d18:1/24:1 [430]). Moreover, an increase in cholesterol was observed that might influence fluidity of the membrane, in addition to other receptor-modifying pathways such as synthesis of ligands. Sequestration of cholesterol in astrocytes, for example, affected APP processing and accumulation of Aβ peptides [431], and serum starvation induced shedding of BACE1, which could be further aggravated by cholesterol efflux mediated by methyl β cyclodextran [432]. TREM2 itself has been shown to act as a regulator of brain cholesterol metabolism, and TREM2-deficient microglia fail to degrade myelin-derived cholesterol [433]. How its polymorphisms or binding of different Aβ variants might interfere with this function has not been addressed to the best of our knowledge.
In sum, this indicates that activation of microglia by Aβ variants can be affected not only directly by genetic variants of the receptors, but also by their impact on receptor processing or on other functions of the receptors that subsequently reflect on Aβ-driven signaling pathways. A deeper understanding of relevance of the single observed variants for the distinct receptor functions on the microglial surface therefore has to be the focus of research in the future to estimate their potential as therapeutic targets or their relevance in individualized medicine.

Secondary Structure and Oligomerization Critically Affect Aβ Neurotoxicity
Different aggregate forms of Aβ that lead to plagues have long been suspected to be the major culprit of AD. However, there is clear evidence indicating that the formation of insoluble aggregates alone is insufficient to trigger typical neuroinflammation and neurodegeneration. First, the formation of insoluble Aβ plaques also occurs during normal aging in many people who do not suffer from AD [434,435]. Second, some patients display typical AD symptoms but show only low plaque formation [436]. Third, transgenic APP mice already show a pronounced plaque formation before the onset of neuroinflammatory events [437]. Finally, therapeutic approaches for plaque removal did not ameliorate cognitive symptoms in human patients [9]. This indicates that at least a second factor has to contribute. Increasing evidence suggests that the formation of soluble oligomers and/or differences in the secondary structure might be the second culprit that triggers the initial inflammatory events. For example, soluble Aβ 1-42 oligomers are highly neurotoxic and can induce strong pro-inflammatory activity in glial cells [438,439]. Next, the Osaka mutation, a familiar form of AD that causes a loss of glutamate at position 22 of Aβ, leads to Aβ peptides with enhanced soluble oligomer formation capability but which are unable to form fibrils, and can therefore not be deposited as insoluble plaques (reviewed in [440]). Studies that injected soluble Aβ 1-42 oligomers into healthy rodents' brains observed strong neurodegenerative effects, including aberrant neuroinflammation, synaptic disruption, neuronal death, and cognitive deficits [441][442][443]. For example, Aβ*56, a distinct species of Aβ oligomers with a molecular weight of 56 kDa, was proposed as a promising candidate to explain the genesis of AD, since the injection of such aggregates could induce strong cognitive decline in rat models [444]. However, recent investigations have questioned the validity of these studies [445]. In addition, other studies could not detect Aβ*56 in tissue [439,446] or CSF samples [438,446] of human AD patients or AD model mice. Next, a number of other studies reported no or only mild pathological effects of Aβ oligomer injection into the brain of animals [447,448]. The interpretation of the divergent results is still problematic. However, technical issues cannot be excluded because these studies used a plethora of different protocols that varied strongly in terms of buffer composition, pH value, aggregation kinetics, Aβ concentration, temperature, aggregation time, and agitation. Numerous studies have demonstrated that even small deviations in the environmental conditions lead to drastic differences in secondary structure, the solubility of the resulting aggregates, and their capabilities to elicit biological effects [449,450].
The secondary and tertiary structure of Aβ is an extremely critical factor for its oligomerization and bio-activity (reviewed in [12]). The secondary structure of Aβ 1-40 and Aβ 1-42 may assume multiple discrete conformations with α-helix or β-sheet conformers, which can undergo rapid changes depending on environmental factors [451,452]. Aβ 1-42 possesses two motifs that are capable of forming β-sheets: the highly hydrophobic core motif KLVFFAE (Aβ [16][17][18][19][20][21][22] [453] and the C-terminal region IIGLMVGGVVIA (Aβ 30-41 ) [451]. β-sheet formation is essential for aggregation, since β-sheet regions of individual peptides self-assemble into cross-linked structures with parallel or antiparallel organization (crossβ patterns) that self-assemble interlinked fibrils and protofibrils [12,454,455]. This selfassembly from monomers in solution is first auto-catalyzed by the formation of nuclei, small aggregates with especially high thermodynamic energy states. These energy levels lead to faster association than dissociation rates of monomers and small oligomers to the nucleus and thus drive further aggregation [456]. This process is enhanced through secondary nucleation where other fibrils or protofibrils in the solution associate with the initial nucleus and accelerate its growth. The nucleation and growth phase finally ends in fibril formation since the fibrillary state is thermodynamically the most stable form of any peptide assembly [456]. In particular, the inflexible hydrophobic C-terminus of Aβ is thought to initiate the transformation from α-helical to β-sheet structure that is needed for the formation of nuclei [12,451]. Accordingly, C-terminal truncation decreases the hydrophobic region and increases the structural flexibility of Aβ, which in turn decreases its propensity to form β-sheets. This may explain why Aβ species with longer C-termini tend to be more amyloidogenic then shorter variants.
Of note, many soluble Aβ variants and their oligomers are present as metastable species with distinct thermodynamic energy levels [457,458]. Some studies have proposed the existence of a α-pleated sheet conformation in a subset of these soluble Aβ oligomers that might set them apart from non-toxic fibrils and are not generated by typical nucleation [459][460][461]. The α-pleated sheet conformation is thought to be structurally similar to β-sheets but may exhibit a distinct biological effect. However, the biological relevance of α-pleated sheet conformations is still debated. A recent study by Shea and colleagues utilized probes designed to catch α-pleated sheet peptides in a transgenic Aβ Caenorhabditis elegans model and in transgenic APP mice, which resulted in a reduction in soluble Aβ oligomers in both models [460].
Taken together, these findings highlight the high importance of keeping track of Aβ's secondary structure during the investigation of its biological effects. This is especially necessary in studies performed with synthetic Aβ peptides. In general, synthesis processes are not performed under physiological conditions and, therefore, can produce peptides with altered conformations [462]. During synthesis, Aβ peptides already aggregate; thus, there is little control over the Aβ species that comprise the final end product [462]. Next, most peptide synthesis processes depend on counter ions such as trifluoracetate or hydrochloride, which are bound to the final synthesis product and can highly influence the solubility, nucleation, and aggregation kinetic of generated Aβ peptides [463,464]. Thus, the use of Aβ peptides derived from different companies and synthesis processes can produce largely different biological effects [113,465], which may additionally vary from batch to batch [456,465]. Various methods have been developed to minimize these aggregation artefacts, e.g., dissolving synthetic Aβ in harsh solvents such as DMSO, HFIP, or NH 4 OH [450,466,467]. However, several studies provided evidence that these treatments can compromise the secondary structure of Aβ [468,469].
Taken together, these findings highlight the need for standardization protocols that improve the comparability and reproducibility of Aβ research. All in vitro experiments should be performed with Aβ obtained from at least two independent sources to validate the observed biological effects and to avoid potential artifacts [113,456,465]. Inexpensive methods such as Thioflavin T aggregation assays or analysis via SDS-PAGE should be used to give some insight regarding the aggregation state of Aβ. Next, all experimental conditions have to be described very carefully because the exact assay buffer composition, pH value, incubation time, assay temperature, and Aβ storage conditions can all influence the outcome of experiments. Finally, it is essential to also report negative or non-conclusive data obtained with different Aβ variants and manufacturers, since these can still provide valuable information for other researchers and may help to elucidate the bigger picture of the still mysterious Aβ.

Concluding Remarks
In summary, the currently available data clearly argue for a significant contribution of non-canonical Aβ variants to the pathogenesis and progression of AD. However, there is still an amazingly large lack of knowledge on the precise contribution of most of these variants. Next, it seems that the influence of gene polymorphisms regarding the recognition of modified or abridged Aβ peptides is even scarcer. Amazingly, only 20 polymorphisms of Aβ interaction partners have so far been clearly associated with AD. Given that for TREM2 alone nearly 200 single nucleotide exchanges have been reported, and that at least 16 other interaction partners exist, presumably thousands of additional SNPs still await a careful analysis of their impact on receptor interaction, proteolytic processing, and shedding. Next, there is clear evidence that microglia, astrocytes, and oligodendrocytes are key players in the production and release of non-canonical Aβ variants. Therefore, the interplay of these cell types with neurons has to be better characterized. Finally, there is a clear need for standardized protocols in in vitro studies and more information on the Aβ peptide composition in in vivo studies. The current literature already provides ample evidence that chemical modifications, extracellular environment, and precise quantities of different amyloid species have a strong influence on the aggregation and bio-activity of Aβ variants. In a recent study, we even showed that supposedly identical Aβ peptides are highly subjectable to solvent-and manufacturer-dependent effects [113]. For the sake of reproducibility, it is thus of uttermost importance to work with more than one peptide in all in vitro experiments, to very precisely report all experimental conditions, and to include data on the secondary structure, 3D conformation, and aggregation kinetics [113,456,465]. In vivo studies urgently need to gather more information on the precise Aβ variant composition, the existing oligomers, and their 3D structure. This will help to identify the key variants and their structural requirements for a given physiological effect.