**4. Pathogenesis**

Pathogenesis is defined as the series of events and mechanisms that result in the clinical manifestation of a disease. It encompasses the aetiological agent, its entry into the host and the subsequent host responses to that agent over time. In general the greater the 'dose' of an aetiological agent, the more rapid will be the disease onset and severity. Host resistance or compensatory mechanisms will act to maintain homeostasis and so delay the disease onset and its impact, but an over exuberant or aberrant host response could also be a major factor in the clinical manifestation of the disease itself. Neurodegenerative diseases tend to be late-onset, around 75 years in the case of AD, with preclinical or prodromal periods of extended, but unknown, lengths. This type of disease history presents a number of pathogenic possibilities. In one scenario the aetiological agent(s) are extremely subtle and require a long time period to overcome host resistance. Alternatively host resistance may be entirely capable of dealing with the neurodegenerative agent until it fails due to old age. In the latter case a much shorter prodromal period would be expected. A variation on the latter theme is that rather than a foreign agent inducing the pathogenic process, it is a physiological process that goes awry because suitable checks and balances fail with advancing age. If the latter proved to be the case with AD then treatment strategies might be plagued by unwanted side effects.

In the introduction we discussed that there are two major forms of AD, monogenic and sporadic. These types of AD are sometimes referred to early-onset (EOAD) and late onset (LOAD), with 60 years of age often given as the arbitrary cut off. With reference to monogenic forms of AD we might think of mutations as an increased 'dose' (severity) of an aetiological agent and monogenic AD can present as young as 20 years of age. In fact, very rarely, AD-causing mutations are actually gene multiplications with direct gene dosage effects.

Sporadic forms of a disease are defined by having no familial or geographic clustering but this term is slightly misleading because there is certainly a genetic component in these common forms of AD. Epidemiological studies suggest that AD sufferers have a 2.5 fold greater likelihood of family history of the disease (Sutherland *et al.*, 2011b). In our 'dose' analogy above the late-onset nature of these common forms of AD would be consistent with common genetic variants, conferring only slight alterations to protein function or expression, being the causative agent. We also generally assume that the genetic component will be multifaceted with both additive and interactive relationships with environmental exposures. The latter refers to a scenario where a potential genetic risk factor only modifies disease risk if that individual has been exposed to a certain environmental stress. We will return to the discussion of AD genetics shortly but it is useful at this stage to introduce the amyloid precursor protein (APP) and its metabolite A.

#### **4.1 APP metabolism**

400 Neuroscience – Dealing with Frontiers

One of the major applications of NGS is full-length mRNA sequencing called RNA-Seq. Unlike microarrays RNA-Seq provides a digital readout of all transcripts including those that are lowly expressed and it does not rely on prior knowledge of the genome for probe design. Most current preparatory methods for RNA-Seq continue to use a PolyA fraction (mRNA) but total RNA methods, which require the deletion of the abundant architectural RNA species (ribosomal and transfer RNA), are improving. The latter allows the full

The term proteome describes the full complement of proteins, including post–translational variants, produced in a particular cell or tissue (Wilkins *et al.,* 1996). For many biologists, proteins remain the key functional entities that can only be approximated by transcriptomic analyses. Certainly there is not necessarily a linear relationship between mRNA and proteins levels. In a generic proteomic analysis, the lysate would be separated by twodimensional gel electrophoresis and the 'spots' of interest excised, digested, and the resultant peptide fragments subjected to mass spectroscopy (MS). Individual spectra are compared to databases to derive peptide identity and by computation, the likely parent protein. This process is also referred to peptide mass fingerprinting. It is now more common to use tandem MS where a specific peptide is further fragmented and fragments subjected to MS (peptide fragmentation fingerprinting). Proteomics can be made semi-quantitative by spiking in stable isotopes as reporter ions allowing the relative abundance of the peptides in

Pathogenesis is defined as the series of events and mechanisms that result in the clinical manifestation of a disease. It encompasses the aetiological agent, its entry into the host and the subsequent host responses to that agent over time. In general the greater the 'dose' of an aetiological agent, the more rapid will be the disease onset and severity. Host resistance or compensatory mechanisms will act to maintain homeostasis and so delay the disease onset and its impact, but an over exuberant or aberrant host response could also be a major factor in the clinical manifestation of the disease itself. Neurodegenerative diseases tend to be late-onset, around 75 years in the case of AD, with preclinical or prodromal periods of extended, but unknown, lengths. This type of disease history presents a number of pathogenic possibilities. In one scenario the aetiological agent(s) are extremely subtle and require a long time period to overcome host resistance. Alternatively host resistance may be entirely capable of dealing with the neurodegenerative agent until it fails due to old age. In the latter case a much shorter prodromal period would be expected. A variation on the latter theme is that rather than a foreign agent inducing the pathogenic process, it is a physiological process that goes awry because suitable checks and balances fail with advancing age. If the latter proved to be the case with AD then

In the introduction we discussed that there are two major forms of AD, monogenic and sporadic. These types of AD are sometimes referred to early-onset (EOAD) and late onset

repertoire of both coding and non-coding RNA to be quantified.

treatment strategies might be plagued by unwanted side effects.

**3.5 Proteomics** 

the overall spectra to be calculated.

**4. Pathogenesis** 

The A peptide was initially purified from amyloid-containing AD brain tissue (Glenner and Wong, 1984) and the cored plaques of AD and Down syndrome patients (Masters *et al.,* 1985). The parent protein, APP, was then isolated from a human brain cDNA library (Kang *et al.,* 1987). The predicted 695 amino acid long protein with a single transmembrane domain was described as being similar to the prion protein, a hypothesised neuronal surface receptor. There are three major APP isoforms, 695, 751 and 770 amino acids in length with APP695 being the most common in neural tissue (Yoshikai *et al.,* 1990).

Following translation the APP protein is retained in the secretory pathway, and is transported to the cell membrane. The protein is subsequently degraded via proteolytic cleavage by either -, and -secretases or alternatively - and -secretases at the cell membrane or following the endocytosis of APP as part of normal membrane turnover (LaFerla *et al.,* 2007). In addition a variable fraction of APP undergoes post-translational proteolytic cleavage within the secretory pathway. The -secretase is a multi-unit enzyme, that includes either the presenilin 1 (PS1) or presenilin 2 (PS2) proteins and it cleaves APP (and other proteins such as Notch) within its transmembrane domain. -secretase activity is carried out by a family of proteins (including ADAM 9) and this cleavage results in the secretion of an extracellular fragment of APP (sAPP-α) and the retention of a C-terminal fragment (CTF) of 83 amino acids (Fig. 2). sAPP appears to act as a neuroprotectant, and has neurotrophic effects on synaptic plasticity (Postina, 2008). Alternatively, and mutually exclusive to -secretase cleavage, the -secretase enzyme (BACE1, also called Asp2, memapsin 2, is the major form in the brain) cleaves APP at variable sites about 16 amino acids proximal to the -secretase site (Vassar *et al.,* 2009). The combination of -secretase and -secretase cleavages releases a variety of peptides that are collectively called A (1,2,3 to 39- 43) (Fig. 2).

Alzheimer's Disease: Approaches to Pathogenesis in the Genomic Age 403

Early linkage studies of multiple large kindreds mapped a common genetic defect to the *APP* region of C21 (St George-Hyslop, 1987 #8988) although *APP* did not initially appear to be involved (Tanzi *et al.*, 1987a; Tanzi *et al.*, 1987b). Missense mutations in *APP* was eventually described in 1991(Goate *et al.,* 1991) while the predicted duplications in the APP gene, given the gene dosage hypothesis for Down syndrome, were only discovered later (Rovelet-Lecrux *et al.,* 2006). These cases also showed cerebral amyloid angiopathy, a disease entity that can occur independently of AD but is present in 80% of sporadic AD cases (Ellis *et al.,* 1996). APP mutations themselves are very rare and it is the PSEN1 gene (encoding presenilin 1) where the majority of AD mutations have been detected (Sherrington, 1995 #674). Mutations in the PSEN2 gene were also found in 1995 (Levy-Lahad *et al.,* 1995) but these are also relatively uncommon (Bekris *et al.,* 2010). In comparison to *PSEN1* mutation carriers the PSEN 2 cases had a more variable phenotype, a later onset of disease and a reduced penetrance, but the ratio of Aβ 1-42:Aβ 1-40 is increased in all monogenic AD brains. One of the PSEN1 families (N141I mutation) was descended from Volga Germans who had immigrated to the USA. It was later suggested that these individuals were likely to originate from the same region of Germany as Auguste D (Yu, 2010 #8999) and perhaps she had the

There is almost irrefutable proof that monogenic forms of AD result from the aberrant production or metabolism of the A peptides and these rare forms are, apart from the earlier age at onset, phenotypically very similar to the sporadic forms (Shepherd *et al.,* 2009). This similarity was the major driver for Hardy and Higgins to propose the amyloid cascade hypothesis for the pathogenesis of sporadic AD in 1992. This hypothesis suggested that the neurotoxic A sets up a cascade of events in adjacent neurons resulting in NFT formation and neuronal death. How A actually precipitates these events was not articulated but most interpreted the hypothesis as suggesting that it was A fibrils in the form of plaques that were the neurotoxic entity. Later amendments to the hypothesis suggested that it was more likely to be A1-42 oligomers rather than fibrils but whether these interact directly with the neuronal cell membrane or via receptors was still unknown (Hardy and Selkoe, 2002). The cascade hypothesis remains the most popular working hypothesis for AD pathogenesis and is the underlying basis for proposed preclinical diagnostic criteria and the vast majority of treatments under development (Sperling *et al.,* 2011). Nevertheless, it is not quite unanimously accepted. Distracters, in particular, point to the fact that NFTs rather than amyloid are initially deposited in the memory-associated hippocampus and that the regional progression of NFTs is a better correlate of disease symptoms and severity (Braak and Braak, 1991). Furthermore the latest GWAs also hint at A-independent mechanisms for

The monogenic forms of AD have also allowed the production of animal models, the main research workhorse towards understanding pathogenesis. A brief synopsis of the mice models is given below but the area has been extensively reviewed by Gotz and Ittner (Gotz and Ittner, 2008) including a thorough consideration of how invertebrate models such the fruitfly (*Drosophila melogaster*) and the worm, *Caenorhabditis elegans*, have also impacted on

same mutation.

AD research.

**4.3 Amyloid cascade hypothesis** 

sporadic forms of the disease (discussed below).

Fig. 2. The structure of the amyloid precursor protein (APP770). This is a schematic diagram of the longest APP isoform (APP770). The -, -and -secretase sites are shown. APP undergoes physiological cleavage through two alternative pathways; -secretase or secretase. The latter pathway involves the cleavage of APP at and sites and results in the production of various A peptides. Cleavage at the site prohibits the formation of the A peptides. The hydrophobic 29-40(2) segment of the A peptide is part of the transmembrane domain of APP and is likely to confer both aggregative and membrane–binding behaviour to these peptides (Sutherland, 2003, PhD thesis).

The β-secretase pathway is also called the amyloidogenic pathway because the major products A1-40 and A1-42 will rapidly aggregate with themselves *in vitro* and *in vivo* with A1-42 in particular capable of rapid self-assembly. A1-42 is known to be neurotoxic *in vitro* (Lorenzo and Yankner, 1994) and *in vivo* (see animal models below) and thought to be central to AD pathogenesis.

Amyloid structures are composed of pairs of anti-parallel β -strands or β-pleated sheets of A. It was these fibrillar forms of A that were originally presumed to be pathogenic although oligomers are now generally regarded as the pathogenic form of A (Walsh *et al.,* 2002). Interestingly A is produced in proportion to synaptic activity and has the opposite effect to sAPP on synaptic connections (Selkoe, 2002). This could suggest an useful antagonistic action although A is still generally regarded purely as a degradative product. A is normally removed from the brain by the perivascular drainage of interstitial fluid to the cervical lymphatics (Weller *et al.,* 2008) but is also actively secreted into the CSF.

#### **4.2 Monogenic forms**

The first AD family presenting with an apparent Mendelian pattern of inheritance was described in 1932 (Schottky, 1932). In more modern times a family was reported with 51 affected persons in 8 generations (Nee *et al.,* 1983). Down syndrome individuals who are trisomic for chromosome 21 (C21) and have 3 copies of the APP gene also develop A pathology (Heston, 1977). A gene dosage model was proposed for Down syndrome patients as an explanation for early presentation with AD-like pathological changes (Tanzi, 1989).

Early linkage studies of multiple large kindreds mapped a common genetic defect to the *APP* region of C21 (St George-Hyslop, 1987 #8988) although *APP* did not initially appear to be involved (Tanzi *et al.*, 1987a; Tanzi *et al.*, 1987b). Missense mutations in *APP* was eventually described in 1991(Goate *et al.,* 1991) while the predicted duplications in the APP gene, given the gene dosage hypothesis for Down syndrome, were only discovered later (Rovelet-Lecrux *et al.,* 2006). These cases also showed cerebral amyloid angiopathy, a disease entity that can occur independently of AD but is present in 80% of sporadic AD cases (Ellis *et al.,* 1996). APP mutations themselves are very rare and it is the PSEN1 gene (encoding presenilin 1) where the majority of AD mutations have been detected (Sherrington, 1995 #674). Mutations in the PSEN2 gene were also found in 1995 (Levy-Lahad *et al.,* 1995) but these are also relatively uncommon (Bekris *et al.,* 2010). In comparison to *PSEN1* mutation carriers the PSEN 2 cases had a more variable phenotype, a later onset of disease and a reduced penetrance, but the ratio of Aβ 1-42:Aβ 1-40 is increased in all monogenic AD brains.

One of the PSEN1 families (N141I mutation) was descended from Volga Germans who had immigrated to the USA. It was later suggested that these individuals were likely to originate from the same region of Germany as Auguste D (Yu, 2010 #8999) and perhaps she had the same mutation.

#### **4.3 Amyloid cascade hypothesis**

402 Neuroscience – Dealing with Frontiers

Fig. 2. The structure of the amyloid precursor protein (APP770). This is a schematic diagram of the longest APP isoform (APP770). The -, -and -secretase sites are shown. APP undergoes physiological cleavage through two alternative pathways; -secretase or secretase. The latter pathway involves the cleavage of APP at and sites and results in the production of various A peptides. Cleavage at the site prohibits the formation of the A peptides. The hydrophobic 29-40(2) segment of the A peptide is part of the transmembrane domain of APP and is likely to confer both aggregative and membrane–binding behaviour

The β-secretase pathway is also called the amyloidogenic pathway because the major products A1-40 and A1-42 will rapidly aggregate with themselves *in vitro* and *in vivo* with A1-42 in particular capable of rapid self-assembly. A1-42 is known to be neurotoxic *in vitro* (Lorenzo and Yankner, 1994) and *in vivo* (see animal models below) and thought to be

Amyloid structures are composed of pairs of anti-parallel β -strands or β-pleated sheets of A. It was these fibrillar forms of A that were originally presumed to be pathogenic although oligomers are now generally regarded as the pathogenic form of A (Walsh *et al.,* 2002). Interestingly A is produced in proportion to synaptic activity and has the opposite effect to sAPP on synaptic connections (Selkoe, 2002). This could suggest an useful antagonistic action although A is still generally regarded purely as a degradative product. A is normally removed from the brain by the perivascular drainage of interstitial fluid to

The first AD family presenting with an apparent Mendelian pattern of inheritance was described in 1932 (Schottky, 1932). In more modern times a family was reported with 51 affected persons in 8 generations (Nee *et al.,* 1983). Down syndrome individuals who are trisomic for chromosome 21 (C21) and have 3 copies of the APP gene also develop A pathology (Heston, 1977). A gene dosage model was proposed for Down syndrome patients as an explanation for early presentation with AD-like pathological changes (Tanzi, 1989).

the cervical lymphatics (Weller *et al.,* 2008) but is also actively secreted into the CSF.

to these peptides (Sutherland, 2003, PhD thesis).

central to AD pathogenesis.

**4.2 Monogenic forms** 

There is almost irrefutable proof that monogenic forms of AD result from the aberrant production or metabolism of the A peptides and these rare forms are, apart from the earlier age at onset, phenotypically very similar to the sporadic forms (Shepherd *et al.,* 2009). This similarity was the major driver for Hardy and Higgins to propose the amyloid cascade hypothesis for the pathogenesis of sporadic AD in 1992. This hypothesis suggested that the neurotoxic A sets up a cascade of events in adjacent neurons resulting in NFT formation and neuronal death. How A actually precipitates these events was not articulated but most interpreted the hypothesis as suggesting that it was A fibrils in the form of plaques that were the neurotoxic entity. Later amendments to the hypothesis suggested that it was more likely to be A1-42 oligomers rather than fibrils but whether these interact directly with the neuronal cell membrane or via receptors was still unknown (Hardy and Selkoe, 2002). The cascade hypothesis remains the most popular working hypothesis for AD pathogenesis and is the underlying basis for proposed preclinical diagnostic criteria and the vast majority of treatments under development (Sperling *et al.,* 2011). Nevertheless, it is not quite unanimously accepted. Distracters, in particular, point to the fact that NFTs rather than amyloid are initially deposited in the memory-associated hippocampus and that the regional progression of NFTs is a better correlate of disease symptoms and severity (Braak and Braak, 1991). Furthermore the latest GWAs also hint at A-independent mechanisms for sporadic forms of the disease (discussed below).

The monogenic forms of AD have also allowed the production of animal models, the main research workhorse towards understanding pathogenesis. A brief synopsis of the mice models is given below but the area has been extensively reviewed by Gotz and Ittner (Gotz and Ittner, 2008) including a thorough consideration of how invertebrate models such the fruitfly (*Drosophila melogaster*) and the worm, *Caenorhabditis elegans*, have also impacted on AD research.

Alzheimer's Disease: Approaches to Pathogenesis in the Genomic Age 405

hyperphosphorylation or truncation as the mechanism. It seemed that tau was sensitising

Excitotoxicity in the brain can result from overactivation of N-methyl-D-aspartate (NMDA) receptors (NRs) and it has been postulated that A causes excitotoxicity through a NRmediated mechanism (Snyder *et al.,* 2005). Under normal conditions the phosphorylation of these NRs by kinases such as Fyn facilitate interactions with post-synaptic density protein 95, the necessary conduit in the perpetuation of the excitatory signal. Interestingly increased Fyn expression was known to exacerbate toxicity in APP transgenic mice (Chin *et al.,* 2005). These factors all appeared unrelated until Ittner and colleagues showed in a dysfunctional tau mouse model, that when the tau-mediated dendritic transport of Fyn was prevented, A-mediated excitotoxicity and premature lethality was attenuated (Ittner *et al.,* 2010).

It remains unclear how A actually decreases the threshold for glutamate excitotoxicity but it is not thought to be through direct binding to the NRs (Snyder *et al.,* 2005). Nevertheless the NR antagonist, memantine is considered to act by blocking A-neurotoxicity (Miguel-

There are very few factors, genetic or environmental, that have reproducibly been shown to modify the risk for sporadic forms of AD. Age, family history, female gender, low education, head injury and type II diabetes seem to increase risk. While long-term antiinflammatory use and performing mental and physical activity seems to be protective. This area has been recently reviewed (Sutherland *et al.*, 2011b). The finding of AD prevalence being inversely associated with education led to the idea of non-demented aged individuals having a 'cognitive reserve' (Stern *et al.,* 1994). This hypothesis suggests that the more our brains are utilised the better they are able to withstand or perhaps more accurately compensate for, increasing AD pathology. Nevertheless, apart from ageing itself, all these

One notable exception here is the possession of the *APOE* ε4 allele which could account for as much as 50% of the attributed risk in AD (Ashford, 2004). This effect was first described in 1993 (Corder *et al.,* 1993) but it is still not known how this common variant actually modifies disease risk. ApoE is the major apolipoprotein of the brain and its primary role is the delivery of lipids and particularly cholesterol to neurons from astrocytes. There are three major protein isoforms (ε2/3/4) based on their respective positions when separated by isoelectric focussing (IEF) (Zellner *et al.,* 2009). The variants are generated by two nonsynonymous single nucleotide polymorphisms (SNPs) in exon 3 of the gene, rs429358 and rs7412. Alternative cytosine or thymine bases at these sites leads to either arginine or cysteine at positions 112 and 158 in apoE protein, respectively. The apoE 4 isoform has arginine at both positions and varies from the common 3 isoform in both its binding affinities (Weisgraber *et al.,* 1982) and degradation properties (Fukumoto *et al.,* 2003; Riddell *et al.,* 2008). The risk of sporadic AD increases with the dose of the APOE 4 allele; heterozygotes are at a two-fold higher risk but 4 homozygotes are at a greater than 12 fold risk (Corder *et al.,* 1993). There is a similar effect on the age at disease onset with an estimated decrease of 7-9 years per allele (Chapman *et al.,* 2001). However, possession of this allele is neither essential nor sufficient for AD. ApoE ε4 has an increased binding

the mice to A-mediated over excitation and potential excitotoxicity.

Hidalgo *et al.,* 2002).

**4.4 Sporadic forms and the APOE 4 genotype** 

factors have been shown to have very minor effects on AD risk.

The transgenic expression of human mutant APP in mice (Games *et al.,* 1995) caused plaques and inflammatory responses but not tangles or any substantial neuronal loss. These mice were subsequently crossed with a variety of gene knockout strains including an apolipoprotein E (APOE) null mouse where a dramatic reduction in A load was seen (Bales *et al.,* 1997). As we will discuss shortly, the possession of the 4 variant of the APOE gene is a leading risk factor for sporadic AD. Holtzman and colleagues were then able to show in these mice that the additional expression of human APOE 4 produced a 10-fold greater increase in A deposits than APOE 3 (Holtzman *et al.,* 2000). A double transgenic mouse was subsequently developed with both APP and PSEN1 mutant forms and this showed both an acceleration of A deposition and an increased quantity of A load over either parent mutant mouse (Holcomb *et al.,* 1998).

A-orientated studies have dominated the AD field but there has always been a subcommunity of researchers that have investigated the role of tau. The first tau transgenic mouse was produced in 1995 and showed that exogenous tau was redistributed away from the axon and hyperphosphorylated in a 'pre-tangle' state (Gotz *et al.,* 1995). Although no mutations in the tau gene (MAPT) have been found in AD patients, tau mutations have been described in a familial form of frontotemporal dementia proving that mutant tau is sufficient for neurodegeneration (Hutton *et al.,* 1998). The first mutant tau mouse strain was produced in 2000 (Lewis *et al.,* 2000) and then a double mutant APP/tau cross was produced in 2001 with the expectation that this would be an accurate phenocopy of AD (Lewis *et al.,* 2001). This double mutant showed the same A pathology as the APP mutant mouse but greater NFT pathology than the parent mutant tau mouse strain and greater neuronal loss. This suggested a potential interaction between A and tau, a possibility strengthened by the demonstration that the intracerebral injection of Aβ1-42, had the same precipitating effect on tau pathology in a second mutant tau strain (Gotz *et al.,* 2001)

It was with the generation of a triple transgenic mouse (APP /tau/PSEN1) that a model that closely recapitulated human AD pathology became available (Oddo *et al.,* 2003). Furthermore A deposition preceded tangle formation in this model supporting the amyloid cascade hypothesis.

The cascade hypothesis does not suggest a direct interaction between tau and A but rather that an extracellular A build up eventually results in altered neuronal tau kinase and phosphatase activity precipitating tau hyperphosphorylation (Hardy and Selkoe, 2002). However support for such a direct interaction came from *in vitro* studies where A in the presence of tau forms fibrillar aggregates containing both molecules (Giaccone *et al.,* 1996). Of course such an interaction would require the intraneuronal build up of A, a hypothesis that remains largely left field despite reasonable evidence supporting it (D'Andrea *et al.,* 2001; LaFerla *et al.,* 2007; Gouras *et al.,* 2010).

A particularly interesting finding was that 'neurotoxic' A failed to cause degeneration of neuronal cultures from tau null mice (Rapoport *et al.,* 2002). In 2007 a similar effect was demonstrated in a mouse over-expressing human APP where behavioral deficits were attenuated on a tau null background (Roberson *et al.,* 2007). As tau did not reduce the levels of APP or A the authors surmised that it must 'uncouple Aβ from downstream pathogenic mechanisms'. They were able to rule out A-mediated modifications of tau such as

The transgenic expression of human mutant APP in mice (Games *et al.,* 1995) caused plaques and inflammatory responses but not tangles or any substantial neuronal loss. These mice were subsequently crossed with a variety of gene knockout strains including an apolipoprotein E (APOE) null mouse where a dramatic reduction in A load was seen (Bales *et al.,* 1997). As we will discuss shortly, the possession of the 4 variant of the APOE gene is a leading risk factor for sporadic AD. Holtzman and colleagues were then able to show in these mice that the additional expression of human APOE 4 produced a 10-fold greater increase in A deposits than APOE 3 (Holtzman *et al.,* 2000). A double transgenic mouse was subsequently developed with both APP and PSEN1 mutant forms and this showed both an acceleration of A deposition and an increased quantity of A load over either parent

A-orientated studies have dominated the AD field but there has always been a subcommunity of researchers that have investigated the role of tau. The first tau transgenic mouse was produced in 1995 and showed that exogenous tau was redistributed away from the axon and hyperphosphorylated in a 'pre-tangle' state (Gotz *et al.,* 1995). Although no mutations in the tau gene (MAPT) have been found in AD patients, tau mutations have been described in a familial form of frontotemporal dementia proving that mutant tau is sufficient for neurodegeneration (Hutton *et al.,* 1998). The first mutant tau mouse strain was produced in 2000 (Lewis *et al.,* 2000) and then a double mutant APP/tau cross was produced in 2001 with the expectation that this would be an accurate phenocopy of AD (Lewis *et al.,* 2001). This double mutant showed the same A pathology as the APP mutant mouse but greater NFT pathology than the parent mutant tau mouse strain and greater neuronal loss. This suggested a potential interaction between A and tau, a possibility strengthened by the demonstration that the intracerebral injection of Aβ1-42, had the same precipitating effect on

It was with the generation of a triple transgenic mouse (APP /tau/PSEN1) that a model that closely recapitulated human AD pathology became available (Oddo *et al.,* 2003). Furthermore A deposition preceded tangle formation in this model supporting the amyloid

The cascade hypothesis does not suggest a direct interaction between tau and A but rather that an extracellular A build up eventually results in altered neuronal tau kinase and phosphatase activity precipitating tau hyperphosphorylation (Hardy and Selkoe, 2002). However support for such a direct interaction came from *in vitro* studies where A in the presence of tau forms fibrillar aggregates containing both molecules (Giaccone *et al.,* 1996). Of course such an interaction would require the intraneuronal build up of A, a hypothesis that remains largely left field despite reasonable evidence supporting it (D'Andrea *et al.,*

A particularly interesting finding was that 'neurotoxic' A failed to cause degeneration of neuronal cultures from tau null mice (Rapoport *et al.,* 2002). In 2007 a similar effect was demonstrated in a mouse over-expressing human APP where behavioral deficits were attenuated on a tau null background (Roberson *et al.,* 2007). As tau did not reduce the levels of APP or A the authors surmised that it must 'uncouple Aβ from downstream pathogenic mechanisms'. They were able to rule out A-mediated modifications of tau such as

mutant mouse (Holcomb *et al.,* 1998).

cascade hypothesis.

tau pathology in a second mutant tau strain (Gotz *et al.,* 2001)

2001; LaFerla *et al.,* 2007; Gouras *et al.,* 2010).

hyperphosphorylation or truncation as the mechanism. It seemed that tau was sensitising the mice to A-mediated over excitation and potential excitotoxicity.

Excitotoxicity in the brain can result from overactivation of N-methyl-D-aspartate (NMDA) receptors (NRs) and it has been postulated that A causes excitotoxicity through a NRmediated mechanism (Snyder *et al.,* 2005). Under normal conditions the phosphorylation of these NRs by kinases such as Fyn facilitate interactions with post-synaptic density protein 95, the necessary conduit in the perpetuation of the excitatory signal. Interestingly increased Fyn expression was known to exacerbate toxicity in APP transgenic mice (Chin *et al.,* 2005). These factors all appeared unrelated until Ittner and colleagues showed in a dysfunctional tau mouse model, that when the tau-mediated dendritic transport of Fyn was prevented, A-mediated excitotoxicity and premature lethality was attenuated (Ittner *et al.,* 2010).

It remains unclear how A actually decreases the threshold for glutamate excitotoxicity but it is not thought to be through direct binding to the NRs (Snyder *et al.,* 2005). Nevertheless the NR antagonist, memantine is considered to act by blocking A-neurotoxicity (Miguel-Hidalgo *et al.,* 2002).

#### **4.4 Sporadic forms and the APOE 4 genotype**

There are very few factors, genetic or environmental, that have reproducibly been shown to modify the risk for sporadic forms of AD. Age, family history, female gender, low education, head injury and type II diabetes seem to increase risk. While long-term antiinflammatory use and performing mental and physical activity seems to be protective. This area has been recently reviewed (Sutherland *et al.*, 2011b). The finding of AD prevalence being inversely associated with education led to the idea of non-demented aged individuals having a 'cognitive reserve' (Stern *et al.,* 1994). This hypothesis suggests that the more our brains are utilised the better they are able to withstand or perhaps more accurately compensate for, increasing AD pathology. Nevertheless, apart from ageing itself, all these factors have been shown to have very minor effects on AD risk.

One notable exception here is the possession of the *APOE* ε4 allele which could account for as much as 50% of the attributed risk in AD (Ashford, 2004). This effect was first described in 1993 (Corder *et al.,* 1993) but it is still not known how this common variant actually modifies disease risk. ApoE is the major apolipoprotein of the brain and its primary role is the delivery of lipids and particularly cholesterol to neurons from astrocytes. There are three major protein isoforms (ε2/3/4) based on their respective positions when separated by isoelectric focussing (IEF) (Zellner *et al.,* 2009). The variants are generated by two nonsynonymous single nucleotide polymorphisms (SNPs) in exon 3 of the gene, rs429358 and rs7412. Alternative cytosine or thymine bases at these sites leads to either arginine or cysteine at positions 112 and 158 in apoE protein, respectively. The apoE 4 isoform has arginine at both positions and varies from the common 3 isoform in both its binding affinities (Weisgraber *et al.,* 1982) and degradation properties (Fukumoto *et al.,* 2003; Riddell *et al.,* 2008). The risk of sporadic AD increases with the dose of the APOE 4 allele; heterozygotes are at a two-fold higher risk but 4 homozygotes are at a greater than 12 fold risk (Corder *et al.,* 1993). There is a similar effect on the age at disease onset with an estimated decrease of 7-9 years per allele (Chapman *et al.,* 2001). However, possession of this allele is neither essential nor sufficient for AD. ApoE ε4 has an increased binding

Alzheimer's Disease: Approaches to Pathogenesis in the Genomic Age 407

Two other factors potentially lowering the power of GWAs are not platform-related at all. As discussed above AD pathology is present in 30% or more of our age-matched controls and there are variations in the clinical presentation of AD. In particular it seems highly probable that individuals who develop clinical variations such as posterior cortical atrophy or logopenic aphasia will have different underlying genetic susceptibilities. It is hoped that biomarkers will be able to facilitate more focused comparisons of AD subgroups or allow the dichotomous case-control paradigm to be replaced by an investigations of a continuous variable such as amyloid load using positive electron

Most readers will associate the term 'microarray' with whole genome expression studies. Microarray (expression) studies have been extensively used over the last decade for the analysis of human tissue and animal models. As discussed above, researchers were expecting a relatively high 40 to 100,000 protein-encoding genes in the human genome. This expectation was largely based on the evolutionarily complex human brain. However comparative genomic analyses have now shown that the increase in complexity and specialisation of the human brain is largely derived at the level of transcription (Enard *et al.,* 2002). This includes the utilisation of variable length 3' untranslated regions (Ramskold *et al.,* 2009) and alternatively spliced isoforms (Pan *et al.,* 2008; Wang *et al.,* 2008). This transcriptomic diversity will underlie both brain function and presumably dysfunction, making the transcriptomic analysis of postmortem human brain tissue a seemingly ideal

A meta-analysis of microarray studies in AD utilising brain tissue has not been undertaken but a review of reported finding suggests that these studies have been largely discordant and disappointingly have not provided novel clues about AD pathogenesis (Courtney *et al.,* 2010). These divergent findings may be due to a number of different factors. High RNA quality is the major factor in the subsequent quality of microarray data. Brain tissue pH is the major determinant of RNA quality but unfortunately it is the agonal period that is critical in determining pH, and this is both long in neurodegenerative disease and out of the control of the researcher. In comparison to animal or cell culture models RNA quality is invariably poorer from postmortem brain tissue (Preece and Cairns, 2003). The study of postmortem tissue is also inherently a retrospective analysis. The time component of neurodegenerative diseases means many rounds of tissue insult and compensatory host responses will have occurred, particularly in regions such as the entorhinal cortex in AD. The pathology in the postmortem brain may have little informative value on what precipitated the disease at the transcriptomic level. In other words it becomes impossible to

Variability in microarray studies may also reflect limitations in the technology itself (Sutherland *et al.*, 2011a). A new NGS-based platform promises to address most of these issues and is discussed immediately below, but prior to this let us briefly look at a combination of the two main methodologies discussed above, GWA and expression studies. These can be combined to derive expression quantitative trait loci (eQTL). In a direct experiment approach genomic DNA of a case-control cohort will be assayed by SNP array while expression analysis will be carried out on say, RNA from peripheral leucocytes of the

tomography (PET) (Sutherland *et al.*, 2011b).

experimental paradigm to find pathogenic clues in AD.

**4.5 Transcriptomics and brain tissue** 

delineate disease 'cause' from 'effect'.

efficiency for A (Strittmatter *et al.,* 1993) and can accelerate A fibril formation *in vitro* (Castano *et al.,* 1995). It is therefore generally interpreted that the apoE 4 isoform modifies AD risk by accelerating the development of plaques although it is probably a more complex association between neuronal lipid content and amyloid precursor protein metabolism (Grosgen *et al.,* 2010).

Before the arrival of GWAs, a meta-analysis of single candidate gene studies showed that, apart from APOE, very few other gene loci modified AD risk and their effects were all very modest (odds ratios ~1.25) (Bertram *et al.,* 2008). The first GWAs in 2007 essentially confirmed these findings (Grupe *et al.,* 2007; Reiman *et al.,* 2007)

Subsequent GWAs would concentrate on defining the non-*APOE* genetic component of AD and in 2009 two independent studies reported an association with a second apolipoproteinencoding gene, CLU, which encodes clusterin or apoJ and a gene called PiCALM (which encodes phosphatidylinositol binding clathrin assembly protein) (Harold *et al.,* 2009; Lambert *et al.,* 2009). Clusterin is best known as a chaperone protein but as its alternative name suggests it is also found in lipoprotein particles and regulates cholesterol and lipid metabolism in the brain (Nuutinen *et al.,* 2009). A feature of these later GWAs were their two-tiered approach where only SNPs deemed significant in a first cohort were tested in an additional independent cohort and only those replicated in both reported as significant. This strategy reduced the multidimensionality of the study and theoretically the risk of generating false positives. Lambert and colleagues also found an additional association with the CR1 gene, encoding a complement receptor (Lambert *et al.,* 2009).

In 2011 two further GWAs were reported that combined these multi-tier approaches with meta-analyses of additional data sets to further increase their detection sensitivity (Hollingworth *et al.,* 2011; Naj *et al.,* 2011). In combination these 2011 studies confirmed 10 significant loci associated with sporadic AD included the genes previously described (APOE, CLU, PICALM and CR1). One of the co-authors from these studies suggested that these 10 genes implicate three pathways in AD pathogenesis, immune system function, cholesterol metabolism and synaptic cell membrane processes (Morgan, 2011). He further suggested that these pathways appear to be largely A-independent. However A is known to inhibit synaptic activity and A can bind to membranes where it may actually modulate the lipid make-up including cholesterol content (Grimm *et al.,* 2005).

Morgan considers that GWAs have now accounted for up to 50% of genetic risk in sporadic AD (Morgan, 2011). The question remains what has happened to the missing heritability? SNP arrays detect common variants (SNP directly and copy number variants by inference) but there is an alternative hypothesis that a large proportion of AD cases will result from rare variants in many different genes (Pritchard, 2001). Alternatively, Cooper and Shendure argue that it is not the GWAs technology that is limiting detection of common variants but rather our interpretation of the data (Cooper and Shendure, 2011). They maintain that an improvement in 'probability' is required before such studies will reach their maximum detection potential. This means that the analysis needs to be limited to functional entities only to reduce the number of tests carried out. They admit that defining which SNPs are functional (or in close linkage disequilibrium with such SNPs) remains a 'work in progress' but report steady advances in both computational approaches that predict changes in protein structure in non-synonymous SNPs and multiplex experimental approaches that combine mutagenesised libraries, *in vitro* transcription and RNA-Seq.

Two other factors potentially lowering the power of GWAs are not platform-related at all. As discussed above AD pathology is present in 30% or more of our age-matched controls and there are variations in the clinical presentation of AD. In particular it seems highly probable that individuals who develop clinical variations such as posterior cortical atrophy or logopenic aphasia will have different underlying genetic susceptibilities. It is hoped that biomarkers will be able to facilitate more focused comparisons of AD subgroups or allow the dichotomous case-control paradigm to be replaced by an investigations of a continuous variable such as amyloid load using positive electron tomography (PET) (Sutherland *et al.*, 2011b).

#### **4.5 Transcriptomics and brain tissue**

406 Neuroscience – Dealing with Frontiers

efficiency for A (Strittmatter *et al.,* 1993) and can accelerate A fibril formation *in vitro* (Castano *et al.,* 1995). It is therefore generally interpreted that the apoE 4 isoform modifies AD risk by accelerating the development of plaques although it is probably a more complex association between neuronal lipid content and amyloid precursor protein

Before the arrival of GWAs, a meta-analysis of single candidate gene studies showed that, apart from APOE, very few other gene loci modified AD risk and their effects were all very modest (odds ratios ~1.25) (Bertram *et al.,* 2008). The first GWAs in 2007 essentially

Subsequent GWAs would concentrate on defining the non-*APOE* genetic component of AD and in 2009 two independent studies reported an association with a second apolipoproteinencoding gene, CLU, which encodes clusterin or apoJ and a gene called PiCALM (which encodes phosphatidylinositol binding clathrin assembly protein) (Harold *et al.,* 2009; Lambert *et al.,* 2009). Clusterin is best known as a chaperone protein but as its alternative name suggests it is also found in lipoprotein particles and regulates cholesterol and lipid metabolism in the brain (Nuutinen *et al.,* 2009). A feature of these later GWAs were their two-tiered approach where only SNPs deemed significant in a first cohort were tested in an additional independent cohort and only those replicated in both reported as significant. This strategy reduced the multidimensionality of the study and theoretically the risk of generating false positives. Lambert and colleagues also found an additional association with

In 2011 two further GWAs were reported that combined these multi-tier approaches with meta-analyses of additional data sets to further increase their detection sensitivity (Hollingworth *et al.,* 2011; Naj *et al.,* 2011). In combination these 2011 studies confirmed 10 significant loci associated with sporadic AD included the genes previously described (APOE, CLU, PICALM and CR1). One of the co-authors from these studies suggested that these 10 genes implicate three pathways in AD pathogenesis, immune system function, cholesterol metabolism and synaptic cell membrane processes (Morgan, 2011). He further suggested that these pathways appear to be largely A-independent. However A is known to inhibit synaptic activity and A can bind to membranes where it may actually modulate

Morgan considers that GWAs have now accounted for up to 50% of genetic risk in sporadic AD (Morgan, 2011). The question remains what has happened to the missing heritability? SNP arrays detect common variants (SNP directly and copy number variants by inference) but there is an alternative hypothesis that a large proportion of AD cases will result from rare variants in many different genes (Pritchard, 2001). Alternatively, Cooper and Shendure argue that it is not the GWAs technology that is limiting detection of common variants but rather our interpretation of the data (Cooper and Shendure, 2011). They maintain that an improvement in 'probability' is required before such studies will reach their maximum detection potential. This means that the analysis needs to be limited to functional entities only to reduce the number of tests carried out. They admit that defining which SNPs are functional (or in close linkage disequilibrium with such SNPs) remains a 'work in progress' but report steady advances in both computational approaches that predict changes in protein structure in non-synonymous SNPs and multiplex experimental approaches that

confirmed these findings (Grupe *et al.,* 2007; Reiman *et al.,* 2007)

the CR1 gene, encoding a complement receptor (Lambert *et al.,* 2009).

the lipid make-up including cholesterol content (Grimm *et al.,* 2005).

combine mutagenesised libraries, *in vitro* transcription and RNA-Seq.

metabolism (Grosgen *et al.,* 2010).

Most readers will associate the term 'microarray' with whole genome expression studies. Microarray (expression) studies have been extensively used over the last decade for the analysis of human tissue and animal models. As discussed above, researchers were expecting a relatively high 40 to 100,000 protein-encoding genes in the human genome. This expectation was largely based on the evolutionarily complex human brain. However comparative genomic analyses have now shown that the increase in complexity and specialisation of the human brain is largely derived at the level of transcription (Enard *et al.,* 2002). This includes the utilisation of variable length 3' untranslated regions (Ramskold *et al.,* 2009) and alternatively spliced isoforms (Pan *et al.,* 2008; Wang *et al.,* 2008). This transcriptomic diversity will underlie both brain function and presumably dysfunction, making the transcriptomic analysis of postmortem human brain tissue a seemingly ideal experimental paradigm to find pathogenic clues in AD.

A meta-analysis of microarray studies in AD utilising brain tissue has not been undertaken but a review of reported finding suggests that these studies have been largely discordant and disappointingly have not provided novel clues about AD pathogenesis (Courtney *et al.,* 2010). These divergent findings may be due to a number of different factors. High RNA quality is the major factor in the subsequent quality of microarray data. Brain tissue pH is the major determinant of RNA quality but unfortunately it is the agonal period that is critical in determining pH, and this is both long in neurodegenerative disease and out of the control of the researcher. In comparison to animal or cell culture models RNA quality is invariably poorer from postmortem brain tissue (Preece and Cairns, 2003). The study of postmortem tissue is also inherently a retrospective analysis. The time component of neurodegenerative diseases means many rounds of tissue insult and compensatory host responses will have occurred, particularly in regions such as the entorhinal cortex in AD. The pathology in the postmortem brain may have little informative value on what precipitated the disease at the transcriptomic level. In other words it becomes impossible to delineate disease 'cause' from 'effect'.

Variability in microarray studies may also reflect limitations in the technology itself (Sutherland *et al.*, 2011a). A new NGS-based platform promises to address most of these issues and is discussed immediately below, but prior to this let us briefly look at a combination of the two main methodologies discussed above, GWA and expression studies. These can be combined to derive expression quantitative trait loci (eQTL). In a direct experiment approach genomic DNA of a case-control cohort will be assayed by SNP array while expression analysis will be carried out on say, RNA from peripheral leucocytes of the

Alzheimer's Disease: Approaches to Pathogenesis in the Genomic Age 409

facilitating plaque development. However there are increases seen in both phosphorylated

A proteomic study that utilised two-dimensional gel electrophoresis (2-DE) found 23 differentially expressed proteins in the CSF of AD patients (Finehout *et al.,* 2007). These included a down-regulation of apoE (but they did not identify A or tau). The sampling of CSF is relatively straightforward but it is not without risk of infection or spinal cord damage. There are also important ethical considerations in consenting dementia patients for such procedures although this issue also extends across the breadth of potential clinical

Researchers and clinicians are therefore keen to find brain-specific (and disease-specific) patterns of expression in serum or plasma samples that are more easily sampled and can be sampled repeatedly during the course of the disease. A disease-specific pattern in peripheral samples could reflect a number of scenarios; pathogenic species have drained or potentially leaked from the brain, there are systemic manifestations of the disease process or the functional consequences of underlying genetic susceptibility can be detected peripherally. As the brain is behind the relatively impermeable blood-brain barrier, leaked metabolites are unlikely to contribute greatly to plasma. Nevertheless the proteins, complement factor H and α-2 macroglobulin have been detected in plasma using by 2-DE and immunochemical assays although their sensitivity (62%) and specificity (60%) is far below those required for a biomarker (Hye *et al.,* 2006). In a recent study the reduced levels of apoE seen in the CSF, have also been seen in plasma of early stage AD sufferers, and particularly in *APOE* 4 carriers. The decrease in apoE levels was inversely correlated with A load seen on PET

As we contemplate the future of research in AD, it is worthwhile reiterating that there are currently no treatments that slow the progression of the disease. At their first clinical presentation AD sufferers will already have significant neuronal loss. Ideally, neuroprotective agents are required but these would still be relatively ineffective if only implemented at the onset of clinical signs. A successful treatment regime is dependent on the co-discovery of a therapeutic target(s) and a preclinical biomarker(s). These, in turn, are

Without doubt the major technological advance in biology is NGS. It seems only a matter of time before the much-anticipated '\$1000 genome' is standard practice in both research and clinical practice (Pareek *et al.,* 2011). Furthermore NGS is about to become third generation sequencing where nucleic acids are sequenced without the need for preparatory PCR amplification of templates, a procedure that potentially introduces biases. These 'third-gen' technologies run at nanoscale proportions and rely on either detecting the exonuclease cleavage of a specific nucleotide from the template or their incorporation into a newly synthesised DNA strand. It is predicted that these new platforms will achieve read lengths of around 1000 base pairs making the subsequent alignment to a reference genome a more

both predicated on gaining a greater understanding of early pathogenic events.

**5.1 Advances in understanding pathogenesis** 

rapid and accurate process.

tau species and total tau in AD patients' CSF.

research.

(Gupta *et al.,* 2011).

**5. Future** 

same individuals. The expression of each individual transcript can be regarded as an independent variable and tested against all the SNPs on the GWAs array. This experiment can detect 'cis' effects of SNPs influencing the expression of their own gene but also 'trans' effects of SNPs that are spatially disparate from the gene of interest.

Alternatively an indirect approach can be used where the expression analysis is carried out in the brain tissue of neuropathologically confirmed cases and controls while the GWAs is in an independent AD cohort from (ideally) the same population. Such an indirect study has been carried out in AD and their major findings were three SNPs that associated significantly with IDE (insulin degrading enzyme) expression levels. This is a very interesting finding as IDE actually degrades A (Kurochkin and Goto, 1994) and a metaanalysis, although not the latest GWAs, suggests that it is associated with sporadic AD (Bertram *et al.,* 2007). Presumably such studies will soon be reported with the partial or complete use of NGS.

#### **4.6 RNA-Seq**

As will be seen in the next section of this chapter NGS will influence nearly all aspects of research on disease pathogenesis. However it may have the greatest impact in trancriptomic analysis (Sutherland *et al.*, 2011a). In comparison to microarrays, RNA-Seq with its linear dynamic range and sensitivity for expression changes in low-abundant transcripts, provides a much more accurate (digital) signal. All transcripts can be confidently assumed to be present regardless of their level of expression. Second, microarrays rely on known genomic sequences for their probe design whereas RNA-Seq, with no such limitation, can detect novel transcripts (Cloonan and Grimmond, 2008). Third, microarrays generally quantify only the total transcripts for each gene but RNA-Seq, with its single base resolution, can detect the exact location of transcription boundaries allowing all transcriptional outputs to be quantified. This includes variants due to alternative promoter usage, splicing patterns and 3' UTR lengths that are unique to the human brain. A recent 'proof of concept' RNA-Seq study compared commercial RNA samples from AD patients with pooled control samples (Twine *et al.,* 2011). One of their major findings was the dysregulation of *APOE* transcription in the temporal lobe of an AD patient. SNPs in and around the APOE gene seemingly associate with AD independently of the ε4 effect, although there is a 7 kb linkage disequilibrium block that covers the entire APOE gene locus (Belbin *et al.,* 2007). It is anticipated that more RNA-Seq studies of postmortem brain tissue will appear in the literature in the near future.

#### **4.7 Proteomics**

The application of proteomics to AD has involved analyses of CSF, plasma and postmortem brain tissue. The driver for this research is to find biomarkers or a proteomic signature to improve AD diagnosis and potentially allow preclinical diagnosis (Zellner *et al.,* 2009).

It has been known for sometime that there are protein-based alterations in the CSF profile of AD patients. Given the role of excess A1-42 in the disease it is perhaps surprisingly to see that this specific peptide is reduced in patient (clinical) analytes (Blennow and Hampel, 2003). This has been explained by the pathogenic retention of A1-42 in the parenchyma

facilitating plaque development. However there are increases seen in both phosphorylated tau species and total tau in AD patients' CSF.

A proteomic study that utilised two-dimensional gel electrophoresis (2-DE) found 23 differentially expressed proteins in the CSF of AD patients (Finehout *et al.,* 2007). These included a down-regulation of apoE (but they did not identify A or tau). The sampling of CSF is relatively straightforward but it is not without risk of infection or spinal cord damage. There are also important ethical considerations in consenting dementia patients for such procedures although this issue also extends across the breadth of potential clinical research.

Researchers and clinicians are therefore keen to find brain-specific (and disease-specific) patterns of expression in serum or plasma samples that are more easily sampled and can be sampled repeatedly during the course of the disease. A disease-specific pattern in peripheral samples could reflect a number of scenarios; pathogenic species have drained or potentially leaked from the brain, there are systemic manifestations of the disease process or the functional consequences of underlying genetic susceptibility can be detected peripherally. As the brain is behind the relatively impermeable blood-brain barrier, leaked metabolites are unlikely to contribute greatly to plasma. Nevertheless the proteins, complement factor H and α-2 macroglobulin have been detected in plasma using by 2-DE and immunochemical assays although their sensitivity (62%) and specificity (60%) is far below those required for a biomarker (Hye *et al.,* 2006). In a recent study the reduced levels of apoE seen in the CSF, have also been seen in plasma of early stage AD sufferers, and particularly in *APOE* 4 carriers. The decrease in apoE levels was inversely correlated with A load seen on PET (Gupta *et al.,* 2011).
