Biomarker Modeling of Alzheimer's Disease

Abstract

Alzheimer's disease (AD) is a slowly progressing disorder in which pathophysiological abnormalities, detectable in vivo by biomarkers, precede overt clinical symptoms by many years to decades. Five AD biomarkers are sufficiently validated to have been incorporated into clinical diagnostic criteria and commonly used in therapeutic trials. Current AD biomarkers fall into 2 categories: biomarkers of amyloid-β plaques and of tau-related neurodegeneration. Three of the 5 are imaging measures and two are cerebrospinal fluid analytes. AD biomarkers do not evolve in an identical manner but rather in a sequential but temporally overlapping manner. Models of the temporal evolution of AD biomarkers can take the form of plots of biomarker severity (degree of abnormality) vs. time. In this review we discuss several time-dependent models of AD which take into consideration varying age of onset (early vs. late) and the influence of aging and co-occurring brain pathologies that commonly arise in the elderly.

Pathological Features of Alzheimer's disease

Two well-known abnormal protein aggregates characterize Alzheimer's disease (AD) pathologically (Hyman et al., 2012). The hallmark of amyloid-β (Aβ) deposits in AD are neuritic plaques and cerebral amyloid angiopathy (CAA). Neuritic plaques are extra cellular and consist of a dense central fibrillar Aβ core with inflammatory cells and dystrophic neurites in its periphery. CAA is also extracellular and consists of fibrillar Aβ deposited in the wall of arterioles in both the leptomeninges and penetrating vessels (Johnson et al., 2007). The second major proteinopathy is aggregated tau protein which are intracellular aggregates of hyper phosphorylated tau in the form of neurofibrillary tangles (NFT). NFTs follow a stereotypic topographic progression pattern first appearing in the brainstem and transentorhinal area then progressing to the hippocampus, to paralimbic and adjacent medial-basal temporal cortex, to cortical association areas, and last to primary sensory-motor and visual areas (Braak and Braak, 1991). Another important pathological feature is neurodegeneration which maps onto NFT distribution topographically (not onto Aβ amyloid distribution) and is characterized macroscopically as atrophy and microscopically as loss of neurons and neuronal processes (Braak and Braak, 1994; Terry et al., 1991).

Clinical-autopsy correlation studies demonstrate a much tighter correlation between NFT and cognitive impairment than between amyloid and cognitive impairment (Bennett et al., 2004; Dickson et al., 1995; Gomez-Isla et al., 1997; Ingelsson et al., 2004). The tightest correlation between cognitive impairment and various autopsy features of AD though seems to be with neurodegeneration (Savva et al., 2009), specifically synapse loss (Terry et al., 1991).

Approximately 30% of cognitively normal elderly subjects have sufficient pathology to meet criteria for AD at autopsy (Knopman et al., 2003; Price and Morris, 1999); however, absence of cognitive symptoms is rarely seen in individuals with severe NFT burden.

Heterogeneity of brain pathology in older adults

The accumulation of brain pathologies seems to be a nearly inevitable consequence of aging; very few elderly individuals have no findings at autopsy and the older the individual the more this holds true (Nelson et al., 2011). The common age-associated brain pathologies are AD (plaques and tangles), medial temporal tangles without plaques, ischemic cerebro-vascular disease (which includes not only macroscopic cortical and subcortical infarctions but also microinfarctions, which seem to be particularly important, and ischemic demyelination), hippocampal sclerosis, alpha synuclein deposits (Lewy bodies), TDP43 inclusions, and agyrophyllic grains (Markesbery et al., 2006; Schneider et al., 2009; Sonnen et al., 2011; White, 2009). These same pathologies are also found in subjects who were cognitively normal at the time of death albeit generally at lower overall pathological burdens than in demented subjects. The most common finding at autopsy in elderly persons who meet clinical and pathological criteria for AD dementia is mixed pathology – plaques and tangles plus one or more of the non-AD pathologies above. Throughout this paper therefore AD is discussed from 2 perspectives: early onset which is a reasonable model of “pure AD”; and late onset which usually consists of AD plus one or more co-occurring pathological processes.

AD pathogenesis: early onset vs. late onset AD

Early Onset AD

AD is commonly categorized clinically as either early onset (onset of clinical symptoms before age 65) or late onset. Early onset AD is uncommon, accounting for a few percent of cases at most. A proportion of early onset AD cases occur in individuals with autosomal dominant mutations in one of three genes; the amyloid precursor protein gene on chromosome 21, the presenilin-1 gene on chromosome 14, or the presenilin-2 gene on chromosome 1 (Goate et al., 1991; Levy-Lahad et al., 1995; Sherrington et al., 1995; St George-Hyslop et al., 1987). Along with Down syndrome, all known autosomal dominant mutations leading to AD influence processing of the amyloid precursor protein (APP) in a manner resulting in 1) increased production of Aβ42 or all Aβ species or 2) increased aggregation/decreased clearance of Aβ (Holtzman et al., 2011). These result in increasing the accumulation of oligomeric and fibrillar Aβ (Scheuner et al., 1996).

It is notable that primary tauopathies lead to forms of fronto-temporal lobar degeneration, cortico-basal degeneration and progressive supranuclear palsy, but never to pathological AD (i.e. not to Aβ accumulation). Animal models in which human amyloid precursor protein with or without presenilin mutations have been inserted into the mouse genome recapitulate the Aβ-linked aspects seen in autosomal dominate AD including increased CSF tau (Maia et al., 2013). Transgenic mouse models of Aβ amyloidosis and cellular evidence indicates that Aβ exacerbates tauopathy (Lewis et al., 2001; Oddo et al., 2004) (perhaps by kindling the auto propagation of NFTs (Clavaguera et al., 2009; de Calignon et al., 2012; Frost et al., 2009; Guo and Lee, 2011)), not the reverse (Oddo et al., 2003)}. The available genetic evidence points to Aβ over production of all Aβ species (APP duplication, APP Swedish mutation), relative Aβ42 versus Aβ40 overproduction (N-terminal APP mutations/presenilin mutations), or enhanced aggregation/decreased clearance (APP mutations within the Aβ domain) as causative in autosomal dominate AD (Holtzman et al., 2011). This has led to the amyloid cascade hypothesis which asserts that dysfunction in the Aβ pathway is the initiating event in the disease (Glenner and Wong, 1984; Hardy and Selkoe, 2002). Soluble Aβ rather than insoluble Aβ is felt to initiate the disease cascade (Hardy, 2009). A simple mechanistic diagram of the amyloid cascade hypothesis is shown below:

Over production/aggregation of Aβ42 ➔ tauopathy ➔ neurodegeneration ➔ clinical symptoms

Late Onset AD

Late onset AD accounts for the overwhelming majority of cases. While most autosomal dominant AD is believed to usually be caused by over-production and subsequent aggregation of Aβ42 (a more fibrillogenic form of Aβ) from the beginning of life, late onset AD may most often be a disease of inadequate Aβ clearance again leading to increased aggregation and accumulation (Mawuenyega et al., 2010). While deterministic genetic mutations for sporadic AD have not been found, genetics nonetheless plays a very important role in risk. The ε4 allele of the apolipoprotein E (APOE) gene is the major known genetic risk factor (Roses, 1994). The ε4 allele of APOE increases the risk of developing AD and also lowers the mean age at onset of the disease in a dose-dependent fashion (Corder et al., 1993; Strittmatter et al., 1993). The major mechanism by which APOE ε4 contributes to AD pathogenesis appears to be by modulating the aggregation and clearance of Aβ peptide (Castellano et al., 2011; Kim et al., 2009) leading to increased deposition (Morris et al., 2010; Vemuri et al., 2010) which implicates this pathway in causation of late onset AD. However, APOE is also implicated in many functions other than Aβ trafficking that may contribute to AD pathogenesis, including regulating brain lipid metabolism, neuronal repair, synaptic function, inflammation, and mitochondrial function (Bu, 2009; Mahley and Huang, 2009). The effect of any specific genetic variant other than APOE4 at the population level is minor. Risk variants including those of CLU, CR1, PICALM, CD33, APP, A673T, ABCA7, EPHA1, TREM 2, SORL1, and BIN1 do point to Aβ but also to alternative pathways including immunologic, inflammatory, neurotropic, endocytosis, cholesterol metabolism, ubiquitination, and tau (Bertram et al., 2008; Griciuc et al., 2013; Guerreiro et al., 2013; Harold et al., 2009; Hollingworth et al., 2011; Jonsson et al., 2013; Lambert et al., 2009; Naj et al., 2011; Seshadri et al., 2010). Thus from a genetics standpoint sporadic AD is complex.

Late onset AD is also more complex than early onset from a pathological perspective. As noted above late onset AD typically occurs along with other age related pathologies. In fact the contribution of pathologies such as hippocampal sclerosis and cerebrovascular disease to dementia seems to increase relative to AD pathology with advancing age beyond 85 years (Nelson et al., 2011). A further complicating factor is the fact that subcortical and medial temporal tauopathy exists at autopsy very commonly by middle age in individuals who have no plaques (Braak and Braak, 1997; Braak and Del Tredici, 2011; Haroutunian et al., 1999; Price and Morris, 1999). For example in autopsies numbering in the thousands, Braak and Braak (Braak and Braak, 1997) find that by the late 70s nearly all individuals (97%) have some tauopathy, while only 17% have Aβ amyloid deposits. The fact that brainstem and medial temporal lobe tau commonly precedes amyloid plaques has led some to suggest that a different pathogenic model should exist for late onset vs. early onset AD.

Role of Aβ in pathogenesis of AD

The initiating event in the molecular cascade that eventually leads to clinical and pathological AD has been controversial for decades. The amyloid cascade hypothesis (Glenner and Wong, 1984; Hardy and Selkoe, 2002) assumes serial causal events initiated by abnormal Aβ production/aggregation. As discussed above, this seems to be a reasonable pathogenic model of early onset AD. An alternative position is that tau hyper-phosphorylation and Aβ elevation are independently arising pathophysiological processes that interact with pathogenic synergy (Duyckaerts, 2011; Duyckaerts et al., 1997; Mesulam, 1999; Small and Duff, 2008) which is particularly germane to late onset AD (Chetelat, 2013; Desikan et al., 2011; Jack et al., 2013a; Knopman et al., 2013a, b). A sequence of pathological events proposed by Price and Morris (Price and Morris, 1999) for late onset AD seems to best explain the fact that while small amounts of medial temporal tauopathy often precede amyloid plaque formation, Aβ seems to drive the progression of the disease. Price and Morris (Price and Morris, 1999) propose that tauopathy develops first, but is confined to subcortical and medial temporal limbic areas, is clinically benign and best thought of as a feature of typical aging. Neocortical Aβ deposits develop later, independently from medial temporal tauopathy. By unknown mechanism(s) distant Aβ aggregation transforms the medial temporal tauopathy leading to spread from medial temporal limbic areas to widespread extension throughout the neocortex. The proposed relationship between cognition and amyloid in late onset AD is not predominantly due to direct neurotoxicity of aggregated Aβ, but rather to its role in aiding the propagation of an antecedent tau related neurodegeneration throughout a topographically characteristic susceptible network (Raj et al., 2012; Seeley et al., 2009). It is tau related neurodegeneration that is ultimately responsible for clinical symptoms (a position reinforced by the poor association between the topographic distribution of amyloid and clinical phenotype in atypical forms of AD (Rabinovici et al., 2008; Wolk et al., 2012)). This view does not conflict with the amyloid cascade hypothesis in that it acknowledges the central role of Aβ in disease pathogenesis. As outlined in a recent review (Jack et al., 2013a) and in the section on AD pathogenesis, while tauopathy and Aβothapy may arise independently, evidence suggests that Aβ drives tau pathology not the reverse. This view also does not imply that therapeutic interventions targeting Aβ in late onset disease are misguided given the essential role Aβ plays in pathogenesis in late onset (as well as early onset) AD.

AD biomarkers

A biomarker is a physiological, biochemical, or anatomic parameter that can be objectively measured as an indicator of normal biologic processes, pathological processes, or responses to a therapeutic intervention. At present five AD biomarkers are well enough established to be used in clinical trials and in modern diagnostic criteria (Albert et al., 2011; Dubois et al., 2010; Jack et al., 2011a; McKhann et al., 2011; Sperling et al., 2011a). Two of these are cerebrospinal fluid (CSF) proteins and three are brain imaging measures. These five biomarkers fall into two major mechanistic categories: First are measures of Aβ deposition: These are CSF Aβ42 which decreases with increasing amyloid plaque load (Bouwman et al., 2009; Fagan et al., 2007; Mattsson et al., 2009; Shaw et al., 2009; Visser et al., 2009), and positron emission tomography (PET) amyloid imaging (Drzezga, 2010; Klunk et al., 2004; Nordberg et al., 2013; Rodrigue et al., 2012; Rowe et al., 2010; Villemagne et al., 2011) (Fig 1a). These two biomarkers of amyloid are highly correlated when measured in the same individuals (Fagan et al., 2006; Jagust et al., 2009; Tolboom et al., 2009; Weigand et al., 2011). The validity of these biomarkers of amyloid plaque deposition has been established by autopsy correlation studies (Clark et al., 2011; Fleisher et al., 2011; Ikonomovic et al., 2008; Sojkova et al., 2011; Strozyk et al., 2003; Tapiola et al., 2009).

Figure 1. Voxel based comparisons of amyloid PET (Fig 1a), FDG PET (Fig 1b) and structural MRI (Fig 1c) illustrate differences between subjects with Alzheimer's disease (n=50) and cognitively normal elderly (n=50).

AD and cognitively normal elderly subjects were from the Mayo Clinic and were matched on age and gender. Fig 1a, amyloid PET maps (thresholded at FWE, P < 0.001 without partial volume correction) illustrate greater retention of Pittsburgh Compound B (PIB) in AD vs cognitively normal elderly in most brain areas; primary sensory motor, visual and the medial temporal lobe are spared. Fig 1b, FDG PET maps (thresholded at FWE, P < 0.001 without partial volume correction) illustrate decreased FDG uptake in the basal temporal, lateral temporal – parietal, lateral pre-frontal, and posterior cingulate-precuneus in AD compared to cognitively normal elderly. This spatial pattern constitutes an “AD-signature” in FDG PET. Fig 1c, structural MRI maps (thresholded at FWE, P < 0.05) illustrate grey matter loss in the medial, basal and lateral temporal, lateral parietal, occipital, insula and precuneus in AD compared to cognitively normal elderly. This spatial pattern constitutes an “AD-signature” in structural MRI. All voxel wise comparisons generated with SPM5. 3D displays by Brain Net Viewer (http://www.nitrc.org/projects/bnv/). The color bar scale indicates the t-test differences between the groups.

The second major AD biomarker category is neurodegeneration, defined as progressive loss of neurons and their processes with a corresponding progressive impairment in neuronal function. The 3 major AD neurodegnerative biomarkers are increased levels of CSF total (t-tau) and phosphorylated (p-tau) tau (Fagan et al., 2009; Mattsson et al., 2009; Shaw et al., 2009; Visser et al., 2009), atrophy on structural MRI (Desikan et al., 2009; Dickerson and Wolk, 2012; Hua et al., 2008; Morra et al., 2008; Morra et al., 2009; Vemuri et al., 2009), and hypo metabolism on FDG PET (Jagust et al., 2010). The validity of these biomarkers of neurodegeneration due to AD is supported by autopsy correlation studies. A caveat is that direct tissue-to-image correlation can be established between tissue stains and in vivo or ex vivo imaging findings, whereas this correlation is more indirect for CSF analytes. Elevations in t-tau (Tapiola et al., 2009) and p-tau (Buerger et al., 2006; Tapiola et al., 2009) correlate with neurofibrillary tangle (NFT) burden at autopsy. Atrophy on MRI correlates with neuron loss (Bobinski et al., 2000; Zarow et al., 2005), Braak NFT stage (Jack et al., 2002; Vemuri et al., 2008; Whitwell et al., 2008), and tau immunostaining burden (Whitwell et al., 2008), and does not correlate well with Aβ load measured by immunohistochemistry (Josephs et al., 2008). Thus MRI is a measure of tau associated neurodegeneration. Antemortem FDG hypometabolism is also correlated to NFT burden and not to plaque burden at autopsy (DeCarli et al., 1992).

An important distinction between biomarkers of Aβ deposition and biomarkers of neurodegeneration concerns their specificity for AD pathophysiology. CSF Aβ42 and amyloid PET are specific for fibrillar Aβ amyloid deposition (in the extracellular space or in vessel walls). In contrast, neurodegenerative biomarkers are not always specific for neurodegeneration due to AD (Jack et al., 2002). In subjects with AD, both FDG PET and MRI follow a modalityspecific topology that is characteristic of AD (Figs 1b and c). An FDG PET or MRI study is transformed into a quantitative biomarker by extracting and summing values from AD-signature regions (as illustrated in Figs 1b and c) using an anatomic atlas that is spatially registered to the subject's imaging study (Senjem et al., 2005). However, atrophy and FDG hypo metabolism caused by non-AD etiologies can overlap spatially with these AD-like topographic patterns. Non-AD etiologies associated with atrophy, FDG hypo metabolism or both that commonly occur in elderly people include cerebro vascular disease, Lewy body disease, hippocampal sclerosis, non-AD degenerative disorders such as Lewy body disease, primary tauopathies, lobar degeneration, Creutzfelt Jakob disease, and perhaps TDP43. A feature of most automated image analysis algorithms that contribute to the non-specificity of MRI and FDG is that the algorithms typically interrogate only a modality-specific AD-like topographic pattern and do not examine areas outside of this predefined topology. Elevation of CSF tau (particularly ptau) seems to be more specific for AD – i.e. it is not elevated in primary tauopathies – however total tau is elevated in Creutzfelt Jakob disease, head trauma and acute stroke. Thus the interpretation of these neurodegenerative biomarkers in empiric studies of elderly subjects is confounded by overlap with co-occurring pathologies that ideally should be distinguished from AD. The importance of this will become evident when discussing modeling of biomarker evolution.

Other potential biomarkers of AD exist at various stages of evaluation. These include other CSF analytes such as VILIP-1 (Tarawneh et al., 2011) diffusion MRI, perfusion MRI, functional MRI (both resting state or task free, and task activation MRI), and tau PET ligands. These are either very new or have not yet demonstrated the test re-test precision or the diagnostic efficacy to be considered a major biomarker. This may change with further development and testing.

Temporal evolution of AD biomarkers

Models of the temporal evolution of AD biomarkers can take the form of plots of biomarker severity (degree of abnormality) vs. time (Fig 2). For imaging biomarkers (Fig 1), movement up the y axis in Fig 2 corresponds to both topographic progression as well as increasing severity in brain areas that are already involved. This type of plot illustrates two key types of relationships in a single display. First is the shape of each biomarker curve as a function of time (i.e. functional form). Biomarkers could increase linearly with time, exponentially, sigmoidally, etc (Fjell et al., 2013b). Different shapes carry different implications concerning the biological mechanisms driving the observed changes in a biomarker with time. The second key relationship concerns the temporal ordering of biomarkers with respect to each other and to the evolution of clinical symptoms. Modeling biomarker trajectories as curves that are time-shifted relative to one another reflects the notion that the different underlying pathophysiologic processes do not evolve simultaneously at identical rates but rather in a temporally ordered manner. A model of biomarker evolution with time depends on both the biology underlying biomarker changes and the sensitivity and specificity of the biomarkers to the biological signals that each responds to. We emphasize that biology (the order or functional form of pathophysiological events) and technology (ability to detect events) are inseparable aspects of the application of biomarkers for human use. For example, it is entirely possible that a highly sensitive biomarker of a later occurring pathological process may become abnormal before a relatively insensitive biomarker of an earlier pathological process.

Figure 2. Temporal evolution of biomarkers.

Fig 2a, biomarker model of pure AD. The horizontal axis is time and the vertical axis severity of biomarker abnormality from completely normal (min) to abnormal (max). The threshold for biomarker detection of pathophysiology is denoted by a horizontal line. The grey area denotes the zone in which abnormal pathophysiology lies below the biomarker detection threshold. Amyloid biomarkers become abnomral first, followed by CSF tau, followed by FDH PET and MRI. Cognitive impairment (green filled area) is the last event in the progression of the disease. A range of cognitive responses are possible that depend on the individual's risk profile. The cognitive response curve is shifted to the left for those with low cognitive reserve and to the right for those with high cognitive reserve. At a given point on the disease time line, a person with high cognitive reserve can be cognitively normal while a low cognitive response person with the same biomarker profile can be impaired. All biomarker curves (as well as cognitive impairment) are configured as sigmoids, but the curves have a progressively steeper slope in the right-hand tail for later-changing biomarkers. The right hand tails of the curves are dashed indicating that biomarker trajectories in end stage dementia are unknown at this time. MCI = mild cognitive impairment. Fig 2b, amyloid-first biomarker model of late onset AD where comorbid pathologies are likely. Figure labels are as in Fig 2a, except here the CSF tau, structural MRI and FDG PET are grouped under the generic neurodegenerative label. This reflects the fact that neurodegenerative biomarkers can be non-specific and that the proportional contribution of various possible etiologies to neurodegeneration cannot be known in vivo. We assume that in elderly individuals tauopathy and in many individuals non-AD neurodegenerative pathologies arise first but lie beneath the detection threshold of biomarkers of neurodegeneration. Aβopathy arises later independently and is the first AD biomarker to become positive. The neurodegeneration curve has a shallow slope initially, but the slope steepens after the onset of amyloidosis. Fig 2c, neurodegeneration-first biomarker model of late onset AD where comorbid pathologies are likely. Figure labels are as in Fig 2a, except here one or more neurodegenerative biomarkers become abnormal first reflecting the onset of tauopathy and in some individuals other neurodegenerative pathologies prior to amyloid. Observing the onset of an abnormal neurodegenerative biomarker prior to an abnormal amyloid biomarker does not alter the role of amyloid as an inducer of neurodegeneration. This is indicated by showing the neurodegeneration curve with a shallow slope initially which steepens after the onset of amyloidosis.

Initial models of the temporal evolution of AD biomarkers were based on observations in elderly subjects spanning the cognitive continuum (Jack et al., 2010a; Jack et al., 2008; Jack et al., 2009; Mormino et al., 2009; Perrin et al., 2009) and were supported by earlier autopsy data (Ingelsson et al., 2004). These initial models proposed that amyloid biomarkers become abnormal first, then biomarkers of tau related neurodegeneration, followed last by overt clinical symptoms. One model took the additional step of proposing not only an ordering scheme but also that the general shape or functional form for biomarker evolution was sigmoidal with time (Jack et al., 2010a). Following publication of these initial biomarker models subsequent empiric studies in both elderly individuals (Buchhave et al., 2012; Forster et al., 2012; Jack et al., 2011b; Jack et al., 2010b; Landau et al., 2012; Lo et al., 2011; Villemagne et al., 2013; Villemagne et al., 2011) and in carriers of autosomal dominant mutations (Bateman et al., 2012; Reiman et al., 2012) found that ordering of biomarkers was generally consistent with that proposed in the models. In addition, empirical data has confirmed that at least one class of biomarkers – amyloid – does follow a sigmoidal shape with time (Caroli and Frisoni, 2010; Fleisher et al., 2012; Jack et al., 2012b; Jack et al., 2013b; Villain et al., 2012; Villemagne et al., 2013). Given that late stage AD subjects have not been thoroughly studied with biomarkers to date, it is not surprising that convincing evidence for plateauing (i.e. sigmoidal shape) has only been found for the earliest changing class of AD biomarkers (amyloid).

The discussion in the previous sections though suggests that different models might exist for early onset vs. late onset AD primarily because cognitive and neurodegenerative biomarker abnormalities due to non –AD pathologies and to aging all may precede or co-occur with AD pathology in the elderly. We propose that 3 different sets of biomarker models should be considered, each of which culminates in the common clinical phenotype of AD dementia.

Pure AD biomarker model – amyloid first

The concept of pathologically pure AD (i.e. in the absence of non-AD age related pathologies) seems appropriate with onset early in life when non-AD age related pathologies are not highly prevalent. This is usually the case in autosomal dominant AD and many APOE4 homozygotes. A model of the temporal ordering of the 5 major AD biomarkers for early onset AD is outlined below and is illustrated in Fig 2a. The temporal ordering scheme in this biomarker model of “pure” AD matches the sequence of molecular events proposed in the amyloid cascade hypothesis (Glenner and Wong, 1984; Hardy and Selkoe, 2002).

1. Aβ 42 over production and aggregation/decreased clearance leads to amyloid plaque formation. CSF Aβ 42 and amyloid PET are the first biomarkers to become abnormal.

2. Aβopathy induces a medial temporal tauopathy and promotes its spread into the neocortex. CSF tau then becomes abnormal. In some cases, Aβopathy also induces synuclein aggregation as well.

3. Tauopathy leads to detectable abnormalities in imaging biomarkers of neurodegeneration -structural MRI and FDG PET.

4. Clinical symptoms follow the MRI and FDG PET abnormalities (Reed et al., 2010; Vemuri et al., 2011).

5. In the early to mid-clinically symptomatic phases of the disease amyloid biomarkers approach a plateau while MRI, FDG PET and tau continue to increase.

6. In the end stage of the disease, we hypothesize that all biomarkers approach a plateau, however this is conjecture. Patients can live for many years after they are no longer able to participate in research studies. In this end stage, patients may survive for prolonged periods unresponsive and bed ridden with little external evidence of further disease progression. Systematic studies of biomarker progression in this stage of the disease have not been done and are difficult to perform. Hence the right hand portions of all biomarker curves in Figure 2 are drawn as dotted lines, indicating lack of knowledge about the terminal phase of the disease.

Late onset AD biomarker models

Modeling late onset AD differs from early onset AD in several important ways. First, AD pathophysiology commonly co-exists with age related non-AD pathologies discussed above (Schneider et al., 2007; Schneider et al., 2009). Second neurodegenerative changes may occur as a function of aging in the absence of any specific disease process, akin to well documented agerelated declines in other organ systems like muscle mass or renal perfusion (Fjell et al., 2013a; Jagust, 2013). Brain systems particularly vulnerable to aging seem to be those sub serving attention/executive function and declarative memory (Jagust, 2013). A candidate substrate for age related declines in in cognitive function, brain volumes and FDG PET uptake in the absence of specific pathologies is synapse loss (Jagust, 2013; Yeoman et al., 2012). Thus unlike early onset AD, late onset AD commonly exists on a background of neurodegenerative changes due to aging, non-AD pathologies, or both (Jack et al., 2010b). Third, a phenomenon that has been recognized by a number of investigators is that biomarkers of neurodegeneration (especially structural MRI and FDG PET) are not specific for AD and may show AD-like patterns of abnormality due to non-AD conditions (Dickerson and Wolk, 2012; Fjell et al., 2013a; Jack et al., 2002; Jack et al., 2013a; Jack et al., 2010a; Jack et al., 2012a; Jack et al., 2010b; Jagust, 2013; Raz et al., 2007; Wirth et al., 2013a; Wirth et al., 2013b). A corollary to this is that validated biomarkers that are specific for non-AD pathologies are lacking. Therefore in any given person, we cannot know the precise pathological etiologies underlying an abnormal structural MRI or FDG PET scan nor the proportional contribution of various possible etiologies to neurodegeneration (Jack et al., 2010b). Realistic interpretation of neurodegenerative biomarkers in the elderly should acknowledge that many possible etiologies and proportional mixes of etiologies may exist for an abnormal neurodegenerative biomarker finding. The reason modeling of AD biomarkers seems to work empirically in the elderly is because AD is the dominant pathology leading to dementia.

While considerable evidence, described earlier under “temporal evolution of AD biomarkers”, supports an amyloid-first biomarker model of late onset AD, evidence has also appeared that supports the idea that cognitive decline and neurodegenerative biomarker abnormalities might precede abnormal amyloid biomarkers in some elderly individuals who later develop AD. This evidence falls into 3 categories. First, autopsy studies indicate that by late 50s 2/3 of the population and by the late 70s essentially everyone has some degree of medial temporal tauopathy very often without amyloid plaques (Braak and Braak, 1997). Therefore tauopathy ought to precede amyloid in many individuals who eventually develop late onset AD (Braak and Braak, 1997; Haroutunian et al., 1999; Price and Morris, 1999). While we currently lack in vivo data on the topographic staging of tau akin to Braak staging (Braak and Braak, 1997), this will likely be remedied in the future by new PET tau ligands (Chien et al., 2013), (Fodero-Tavoletti et al., 2011; Maruyama et al., 2013). Second, FDG PET and functional MRI studies have identified AD-like imaging abnormalities in cognitively normal APOE4 carriers (compared to non-carriers) who were either too young to be amyloid positive (Filippini et al., 2009; Reiman et al., 2004) or who were documented to be amyloid negative (Jagust and Landau, 2012; Sheline et al., 2010). These APOE4 carriers are much more likely to develop AD in the future than non-carriers. While these studies are suggestive they do not prove that neurodegenerative abnormalities can precede amyloidosis in individuals who do develop incident pre-clinical AD. That requires longitudinal documentation of incident amyloid positivity. To that end, a longitudinal study has recently proved proof that MRI and FDG PET can be abnormal prior to documented indicant amyloid positivity in the elderly (Jack et al., 2013c). This study (Jack et al., 2013c) has demonstrated empirically that both “amyloid-first” and “neurodegeneration-first” biomarker profile pathways to pre-clinical AD exist. We therefore describe both amyloid first and neurodegeneration first models of late onset AD below.

The sequence of molecular and biomarker events in our proposed amyloid-first model of late onset AD is illustrated in Fig 2b and outlined below.

1. Neurodegeneration due to tauopathy, non-AD pathologies, aging or a combination first appears in the brainstem and medial temporal lobe but lies beneath the detection threshold of biomarkers of neurodegeneration.

2. Aβ deposition arises later and progresses independently in neocortical association areas. Aβ biomarkers (CSF Aβ42 and amyloid PET) then become positive.

3. Aβopathy transforms slowly progressing medial temporal tauopathy into an aggressive process and induces its spread into the neocortex (Clavaguera et al., 2009; de Calignon et al., 2012; Frost et al., 2009; Guo and Lee, 2011; Price and Morris, 1999). One or more neurodegenerative biomarkers then become abnormal. Because in any older subject we cannot know if AD pathology exits in isolation or is accompanied by co-existing neurodegenerative processes, we group all neurodegeneration biomarkers together under a single generic neurodegeneration heading in this model without attempting to distinguish tau from non-tau related etiologies (which is different from the model of early onset AD where CSF tau and MRI/PET are treated distinctly because non-AD pathology is unlikely to be present). The accelerating effect of amyloid on neurodegeneration is illustrated in the neurodegeneration curve which has a shallow slope initially, but the slope steepens after the onset of amyloidosis.

4. Clinical symptoms follow the onset of abnormalities in neurodegeneration biomarkers. Subjects who are at high risk of cognitive impairment from AD are shown in Figure 2 with a cognitive response curve that is shifted to the left in time (Jack et al., 2013a; Jack et al., 2010a). Such high-risk subjects may harbor more significant co-morbid pathologies, more genetic risk alleles, or have low cognitive reserve (Reed et al., 2010; Rentz et al., 2010; Roe et al., 2008; Vemuri et al., 2011). In contrast, low-risk subjects with fewer co-morbid brain pathologies, a protective genetic profile and high cognitive reserve can co-exist with substantial AD pathophysiology and still maintain normal cognitive function. Thus cognitive response in Figure 2 is illustrated as a zone with low and high risk borders (Jack et al., 2013a). The variance in cognitive response (i.e. the range on the y axis projection) decreases as end stage dementia is approached which reflects the fact that there is relatively little variation in the end stage clinical phenotype.

The sequence of molecular and biomarker events in our proposed neurodegeneration-first model of late onset AD is illustrated in Fig 2c and outlined below.

1. Age related neurodegeneration due to medial temporal tauopathy, non-AD pathologies, aging or a mixture occur first. CSF tau, MRI, FDG PET or any combination of these become abnormal and are denoted as a single generic neurodegeneration biomarker curve.

2. Aβ deposition arises independently in neocortical association areas. CSF Aβ42 and amyloid PET become abnormal.

3. Aβopathy transforms tau related neurodegeneration into an aggressive process and induces its spread throughout the neocortex (Clavaguera et al., 2009; de Calignon et al., 2012; Frost et al., 2009; Guo and Lee, 2011; Price and Morris, 1999). As outlined in the section on the role of Aβ in pathogenesis of AD, evidence that neurodegenerative imaging measures can become abnormal before amyloid biomarkers does not negate the central role of Aβ in late onset AD where Aβ is envisioned as an accelerator of tau related neurodegeneration (rather than an initiator of tau related neurodegeneration as in early onset AD).

4. Neurodegenerative biomarker abnormalities then accelerate.

5. Clinical symptoms follow the progression of neurodegenerative biomarker abnormalities.

Conclusions

AD is a slowly evolving disorder in which pathophysiological abnormalities precede overt clinical symptoms by many years to decades. The only in vivo window into the disease in its long pre-clinical phase is biomarkers which, unlike autopsy, can be sampled repeatedly in individual subjects over time. A thorough understanding of the disease hinges on developing accurate, comprehensive models of AD biomarker evolution. Designing interventional strategies that target the right molecular pathways at an appropriate stage in the disease depends on accurate models of AD biomarker evolution (Fox et al., 2005; Frisoni and Delacourte, 2009; Rinne et al., 2010; Sperling et al., 2011b). This will lead to interventions that may be complex but are correctly tailored to the individual patient – i.e. individualized medicine. Finally, evolving diagnostic criteria (Albert et al., 2011; Dubois et al., 2010; Jack et al., 2012a; McKhann et al., 2011; Sperling et al., 2011a) must incorporate AD biomarkers, but cannot proceed effectively without accurate models of those biomarkers.

In Brief Paragraph.

We intend to familiarize readers with the major biomarkers of Alzheimer's disease and relate them to specific pathophysiological processes. We also outline a framework for temporal evolution of AD biomarkers in the context of both early and late onset disease.