Neuroanatomy of Alzheimer’s Disease

1. The Anatomy of Memory — Hippocampus & Surrounding Cortex

Episodic memory — the form of memory destroyed earliest and most completely in Alzheimer’s disease — is supported by a tightly organised circuit centred on the hippocampal formation and its surrounding allocortex within the medial temporal lobe (MTL). The standard anatomical circuit, mapped in primate by Suzuki and Amaral and in human by Insausti, runs:

The entorhinal cortex (Brodmann area 28) is the obligate gateway: virtually every cortical input that reaches the hippocampus passes through it, and virtually every hippocampal output returns through its deep layers. Layer II of entorhinal cortex projects to dentate gyrus and CA3 via the perforant path; layer III projects directly to CA1.

• Hippocampus proper (cornu Ammonis CA1–CA4 + dentate gyrus + subiculum) — the trisynaptic loop. CA1 pyramidal neurons are highly vulnerable to Aβ-driven excitotoxicity and to ischaemic injury (Sommer’s sector).
• Entorhinal cortex (BA 28) — the first cortical structure to develop neurofibrillary tangles. Layer II stellate cells (the “island cells” of Klink and Alonso) are the earliest neuronal target.
• Perirhinal cortex (BA 35/36) — object/item recognition memory. Brown and Aggleton (Nat Rev Neurosci 2001) made the case that perirhinal supports familiarity judgments distinct from hippocampal recollection.
• Parahippocampal cortex (BA 36/TF/TH) — spatial / contextual memory. Forms part of the “scene-construction” network with retrosplenial cortex.
• Subiculum — the major efferent of the hippocampus, projecting to entorhinal layer V, mammillary bodies (via fornix), and the nucleus accumbens.

The seminal lesion case H.M. (Henry Molaison, operated by William Scoville in 1953; described by Brenda Milner 1957) established that bilateral medial-temporal-lobe resection produces dense anterograde amnesia with intact procedural and working memory. AD recapitulates H.M. molecularly — the entorhinal cortex and CA1 are degraded years before any other cortical region.

2. The Default-Mode Network — Why AD Targets It Preferentially

Resting-state fMRI has revealed a set of cortical regions that are more active when the brain is at rest than during external task performance — the default-mode network (DMN), named by Marcus Raichle (PNAS 2001) and characterised functionally by Buckner, Greicius, and Andrews-Hanna. Its core nodes are:

• Posterior cingulate cortex (PCC) / precuneus — the network hub.
• Medial prefrontal cortex (mPFC) — especially ventral mPFC (BA 10/11/32).
• Inferior parietal lobule / angular gyrus (BA 39).
• Lateral & medial temporal lobe, including the hippocampal formation.
• Retrosplenial cortex (BA 29/30).

The DMN supports self-referential thought, autobiographical memory retrieval, prospection, and theory of mind. Critically, Buckner et al. (J Neurosci 2005, J Neurosci 2009) showed that the spatial topography of amyloid-PET deposition in AD matches the DMN almost one for one. The same regions that are most metabolically active at rest accumulate Aβ first and most densely.

Greicius et al. (PNAS 2004) showed that DMN connectivity, especially between PCC and hippocampus, is reduced in mild AD — one of the earliest functional signatures, detectable before structural atrophy. The precuneus / PCC is consistently the first region to show FDG-PET hypometabolism.

3. Braak Staging of Tau Pathology

In 1991 Heiko Braak and Eva Braak (Acta Neuropathol 1991; Neuropathological stageing of Alzheimer-related changes) examined 83 brains across the spectrum of cognitive status using Gallyas silver staining, and described an invariant six-stage trajectory by which neurofibrillary tangles spread through the brain. The Braak stages remain the single most cited framework in Alzheimer neuropathology: every clinical and biomarker scheme that has followed (NIA-AA, ATN, tau-PET tracers) is calibrated against them.

The Braak hypothesis is mechanistically extended by the tau-spread / prion-like hypothesis (Frost & Diamond, Nat Rev Neurosci 2010; Clavaguera et al., Nat Cell Biol 2009): misfolded tau templates the misfolding of native tau in connected neurons, propagating along axonal projections like a corruption wave following the connectome. This makes Braak stages predictions of network connectivity rather than diffusion: tau jumps synapses, not millimetres.

The clinical lesson: tau topography ≈ clinical phenotype. Atypical AD variants — posterior cortical atrophy (PCA), logopenic aphasia, frontal-variant AD — map onto displaced Braak trajectories that bypass the medial temporal lobe and concentrate tau in parietal, lateral-temporal, or frontal cortex respectively, while still amyloid-positive.

4. Thal Phasing of Amyloid Pathology

Where Braak mapped tau, Dietmar Thal and colleagues mapped amyloid. Thal et al. (Neurology 2002; Phases of Aβ-deposition in the human brain and its relevance for the development of AD) defined a five-phase trajectory of Aβ plaque deposition that runs — strikingly — in the opposite direction from Braak. Tau begins limbic and spreads outward; amyloid begins neocortical and spreads inward and downward.

• Phase 1 — neocortex (frontal, temporal, parietal isocortex; basal portions first).
• Phase 2 — allocortex (entorhinal cortex, hippocampus CA1, insular cortex, cingulate).
• Phase 3 — diencephalic nuclei (thalamus, hypothalamus, striatum).
• Phase 4 — brainstem (substantia nigra, raphe, reticular formation).
• Phase 5 — cerebellum (molecular layer plaques).

That tau and amyloid follow opposite trajectories is one of the central puzzles of AD neuroanatomy: amyloid blankets the cortex without much regional preference for cognition, while tau follows a stereotyped path that tracks symptomatology precisely. The implication — reinforced by the modern era of tau-PET — is that amyloid is necessary but not sufficient: it is a required substrate, but tau is the proximal driver of clinical decline.

The corresponding clinical biomarker schemes (CERAD plaque score; NIA-AA “A” positivity by amyloid PET or CSF Aβ42:Aβ40) effectively dichotomise patients at roughly Thal phase 2–3, where neocortical signal becomes detectable.

5. Hippocampal Atrophy and the MTL Memory System

Larry Squire’s medial-temporal- lobe (MTL) memory system model (Squire & Zola-Morgan, Science 1991; Squire, Stark & Clark, Annu Rev Neurosci 2004) is the framework against which AD memory loss is read:

• Declarative memory (episodic + semantic) is supported by the MTL.
• Non-declarative memory (procedural, priming, classical conditioning) is supported by basal ganglia, cerebellum, and neocortex independent of MTL.
• Within the MTL, the hippocampus binds disparate cortical inputs into a unified episodic representation; the parahippocampal cortices contribute object and contextual detail.
• Memories consolidate from MTL-dependent to neocortically-distributed storage over months to years (Squire’s standard model; Frankland & Bontempi, Nat Rev Neurosci 2005).

In AD this is exactly what is observed: recent (MTL-dependent) memory is destroyed first, while remote (consolidated) memories survive far longer — the Ribot gradient. Procedural memory is preserved until late stages, accounting for the patient who can still play piano while no longer recognising her children.

Volumetric MRI demonstrates the anatomical correlate. Hippocampal volume in early AD is reduced ~15–25% compared with age-matched controls; entorhinal cortex is reduced ~30–40%, often with a greater effect than hippocampus — consistent with entorhinal preceding hippocampus in Braak. The hippocampal atrophy rate in AD is ~3–5%/year (vs ~0.5–1%/year in normal aging) (Jack et al., Neurology 2000+; Frisoni et al., Lancet Neurol 2010).

The Scheltens Medial Temporal Atrophy (MTA) score — a 0–4 visual rating of choroid fissure, temporal horn and hippocampal height on coronal MRI — is the workhorse clinical instrument. Quantitative pipelines (FreeSurfer, FastSurfer, NeuroQuant) provide automated hippocampal segmentation now embedded in the diagnostic workup.

6. Cholinergic System Degeneration

Long before amyloid was sequenced, the cholinergic hypothesis dominated AD biology and was the basis of the first generation of effective drugs. Two papers in the same year cracked it open:

• Davies & Maloney (Lancet 1976): markedly reduced choline acetyltransferase (ChAT) activity in AD cortex and hippocampus.
• Bowen et al. (Brain 1976): the same finding from an independent autopsy series.
• Whitehouse et al. (Science 1982): the source — profound neuronal loss in the nucleus basalis of Meynert.

The basal forebrain cholinergic system consists of four cell groups (Mesulam Ch1–Ch4):

The nucleus basalis of Meynert (Ch4) loses up to ~75% of its neurons in moderate AD. Because Ch4 is the sole source of cortical acetylcholine, its degeneration is the anatomical substrate of the cholinergic deficit. Pre- and post-synaptic cholinergic markers (ChAT, vesicular acetylcholine transporter, M1/M2 receptors) are reduced throughout cortex and hippocampus.

Crucially, cholinergic projections supply the very regions earliest hit by tau (entorhinal, hippocampus, association cortex) — raising the question of whether cholinergic deficit is a cause or consequence of the Braak trajectory. Modern data (Mesulam, Ann Neurol 2013) suggests Ch4 cholinergic neurons are themselves an early site of tau pathology — bringing them close to the locus-coeruleus-first revision below.

7. The Locus Coeruleus and the Earliest Pathology

The most striking modern revision of Braak is the recognition that tau pathology may not start in the entorhinal cortex at all. In a series of methodologically careful papers, Heiko Braak and Kelly Del Tredici (Acta Neuropathol 2011; J Neuropathol Exp Neurol 2011; Brain 2018) re-examined hundreds of brains across the lifespan using highly sensitive AT8 immunohistochemistry — which detects pretangle / soluble hyperphosphorylated tau invisible to silver staining — and found tau pathology in the locus coeruleus (LC) in subjects as young as their twenties, decades before any cortical changes.

The LC is a small (~50,000 neurons per side) pigmented nucleus in the dorsolateral pontine tegmentum, the brain’s sole source of cortical noradrenaline. Its diffuse axonal network reaches every cortical area, the hippocampus, the cerebellum, and the spinal cord. In the Braak/Del Tredici scheme, AD neurofibrillary pathology progresses through six stages beginning before the original Braak I:

Why the LC? Several converging features make it uniquely vulnerable:

• Massive axonal arborisation — a single LC neuron may give rise to ~250,000 cortical varicosities. Maintaining this proteome is metabolically punishing.
• Pigmented (neuromelanin-laden) — oxidation of catecholamines generates reactive oxygen species; iron accumulation amplifies oxidative stress.
• Anatomical access — LC dendrites lie close to the fourth-ventricle CSF and are well placed to take up extracellular tau seeds.
• Connectome topology — LC projects to entorhinal cortex; transsynaptic tau spread from LC to entorhinal would explain why entorhinal is the next station.

The functional consequences are non-trivial: noradrenergic loss from LC degeneration compromises cortical arousal, attention, sleep–wake regulation, and the neuro-immune brake that NA imposes on microglial activation (Heneka et al., Nat Rev Neurosci 2010). LC loss is therefore mechanistically tied to the neuroinflammatory phenotype of AD.

In vivo imaging of LC has caught up: neuromelanin-sensitive MRI sequences (Sasaki, Chen and Mather, Nat Rev Neurosci 2017) now permit measurement of LC integrity in living humans, and reduced LC signal correlates with prodromal AD, REM-sleep behaviour disorder, and Parkinson’s disease. The LC is becoming a target for early-detection and even early-intervention trials (e.g., atomoxetine repurposing).

8. Cortical Atrophy Patterns

On structural MRI, AD has a recognisable temporoparietal-predominant atrophy pattern:

• Hippocampus & entorhinal cortex — atrophied earliest and most severely.
• Inferior temporal / fusiform / angular gyrus — major cortical loss; corresponds to language and visuospatial deficits.
• Posterior cingulate / precuneus — the DMN hub; among the first to show metabolic drop on FDG-PET.
• Lateral parietal cortex — involved in moderate disease.
• Frontal association cortex — later, with executive decline.
• Primary motor & primary visual cortex — relatively spared until late (which is why gait and visual acuity are preserved).

FDG-PET shows the classic “temporoparietal hypometabolism with sparing of sensorimotor cortex” signature, often bilateral but asymmetric, with PCC/precuneus the most reliable early site (Minoshima et al., Ann Neurol 1997; Mosconi, Eur J Nucl Med 2005). The cortical-thinning pattern derived from FreeSurfer cortical-thickness maps — the “AD signature” of Bakkour, Dickerson and colleagues (Cereb Cortex 2009) — provides a continuous biomarker that tracks progression even into preclinical stages.

In atypical variants the topography reorients but the principle survives:

• Posterior cortical atrophy (PCA) — bilateral parieto-occipital atrophy; visuospatial onset (Benson 1988; Crutch et al., Lancet Neurol 2012).
• Logopenic primary progressive aphasia (lvPPA) — left temporoparietal atrophy; phonological loop loss (Gorno-Tempini et al., Neurology 2011).
• Frontal-variant AD — bilateral frontal atrophy; behavioural / dysexecutive presentation, easily mistaken for bvFTD.

In all of these, amyloid is positive throughout the cortex (Thal phase ≥3), while tau localises to the symptomatic region — another argument for tau as the proximal driver and amyloid as the permissive substrate.

9. White Matter Changes

AD is not purely a grey-matter disease. White-matter pathology contributes to network-level dysfunction and is mechanistically intertwined with vascular co-pathology that is the rule in older AD brains.

• Diffusion tensor imaging (DTI) — reduced fractional anisotropy (FA) and increased mean diffusivity in tracts connecting AD-targeted regions: cingulum bundle (fornix, parahippocampal cingulum), uncinate fasciculus, inferior longitudinal fasciculus, genu of corpus callosum. Bozzali et al. (J Neurol Neurosurg Psychiatry 2002); Mielke et al. (Neurology 2009).
• Fornix degeneration — the major efferent of the hippocampus; FA in fornix tracks hippocampal volume and predicts MCI conversion better than either alone.
• Wallerian degeneration — cortical neuron loss is followed by retrograde and anterograde axonal degeneration; much of late white-matter loss is secondary.
• White-matter hyperintensities (WMH) on FLAIR — reflect small-vessel ischaemic disease; a major modifier of cognitive expression of AD pathology.
• Cerebral amyloid angiopathy (CAA) — Aβ deposition in the walls of cortical and leptomeningeal arterioles; present in >80% of AD cases at autopsy. Causes lobar microbleeds, cortical superficial siderosis, and (in severe cases) lobar haemorrhage.

CAA matters clinically because it is the substrate of ARIA (Amyloid-Related Imaging Abnormalities) — the MRI signature of vasogenic oedema (ARIA-E) and microhaemorrhages (ARIA-H) seen during anti-amyloid antibody therapy. ARIA risk is sharply increased in APOE-ε4 homozygotes whose vessels are already amyloid-laden. CAA staging (Vonsattel et al., Ann Neurol 1991) parallels but does not coincide with parenchymal Aβ deposition.

The two-hit vascular hypothesis (Zlokovic, Nat Rev Neurosci 2011) proposes that blood–brain-barrier dysfunction and reduced cerebral blood flow are upstream drivers of the proteinopathy, not merely passive bystanders — tying AD to the broader vascular-cognitive-impairment continuum.

10. The Glymphatic System & Sleep

The brain has no conventional lymphatic system, and how it cleared its interstitial waste was unsettled until 2012. Jeffrey Iliff, Maiken Nedergaard and colleagues described the glymphatic system (Iliff et al., Sci Transl Med 2012; A paravascular pathway facilitates CSF flow through the brain parenchyma and the clearance of interstitial solutes, including amyloid-β):

• CSF enters the brain along peri-arterial Virchow–Robin spaces, driven by arterial pulsations.
• It crosses into the interstitium through astrocytic aquaporin-4 (AQP4) water channels in astrocyte end-feet.
• Bulk flow through the parenchyma carries solutes — including soluble Aβ and tau — out via peri-venous spaces and ultimately to dural lymphatics (Louveau et al., Nature 2015).

The system is sleep-dependent. Xie et al. (Science 2013; Nedergaard lab) showed that during NREM sleep the interstitial space expands ~60%, and Aβ clearance roughly doubles. Slow-wave sleep specifically — the deep oscillation that consolidates declarative memory — is the period of highest glymphatic flow. This places sleep at the intersection of memory consolidation and proteostasis.

Glymphatic dysfunction is now studied in vivo via:

• DTI-ALPS index (Taoka et al., Jpn J Radiol 2017) — diffusivity along the perivascular space; reduced in AD and PD.
• Intrathecal gadolinium tracer studies (Eide and Ringstad) — direct visualisation of CSF turnover.
• Sleep-EEG biomarkers — slow-wave amplitude and spindle–slow-wave coupling, both reduced in preclinical AD.

The glymphatic story closes a circle. The DMN (section 2) accumulates Aβ because it runs hottest at rest; sleep is the period during which the cost of that activity is paid back by clearance; the LC (section 7) is the master regulator of sleep–wake state and itself one of the earliest casualties. AD is thus a disease of failed homeostasis in a brain organ that lives on the edge of its metabolic budget — the anatomical foundation we will build on molecularly in Part III (Amyloid Cascade) and Part IV (Tau), then revisit clinically in Part VI (Diagnosis & Biomarkers). For the broader neuroanatomy backdrop see the Neuroscience course.