Brain diseases

Diseases are natural lesion experiments — what's broken tells you what the part normally did. A taxonomy by mechanism (degeneration, mis-precision, miswiring, vascular, paroxysmal) is more useful than DSM symptom-clusters because it predicts what trains, what slows decline, and what is structurally fixed.

🌿 budding tended 2026-05-10 research disease lesion neurology psychiatry

flowchart LR
  mech[mechanism] --> sx[symptoms]
  sx --> dsm[DSM cluster]
  mech -.- truthful
  dsm -.- pragmatic
  mech --> tx[what helps]
  dsm -. weakly .-> tx

Connected work

• brain structure — what each disease lesion targets
• humans as generators — which dial each disease breaks
• brain memory mgmt — memory diseases in detail
• brain ↔ body axis — diseases that cross the axis

Investigation · rating: medium. Synthesis page; clinical decisions need clinicians, not this.

Status: budding | 2026-05-10 | rating: medium Compress levels: L0 ↓ L1 ↓ L2

L0 — TL;DR (≤5 lines)

Diseases of the brain are natural lesion experiments. Each major class disables a specific component of the system — neurons, synapses, networks, dopamine, vasculature, or the gating that holds the whole thing together. What's lost names what that component did. A mechanism-based taxonomy (degeneration / mis-precision / miswiring / vascular / paroxysmal / autoimmune) predicts what trains, what slows, and what is structurally fixed; the DSM symptom clusters do not.

L1 — Overview

Core question

What does each major brain-disease class teach us about the normal brain — and what is therefore trainable, slowable, or beyond reach?

Why it matters

• Clinical taxonomy (DSM, ICD) is symptom-based and overlapping: depression, anxiety, ADHD, and PTSD share lots of symptoms because the underlying mechanisms overlap, not because the categories carve nature at the joints.
• A mechanism-based taxonomy is rougher but more predictive — it tells you which network, which neurotransmitter, which structural component is implicated, which constrains both research and expectations.
• This is the human analogue of how the swarm reasons about failure modes — a failed lesson tells you what the pipeline was for. Disease is the same logic at biological scale.

Mermaid map (L1)

flowchart LR
  brain[Healthy brain] --> mech{Mechanism}
  mech --> deg[Degenerative — protein aggregation, neuron loss]
  mech --> psy[Psychiatric — mis-precision, network imbalance]
  mech --> dev[Developmental — atypical wiring]
  mech --> vasc[Vascular — infarct, hemorrhage]
  mech --> par[Paroxysmal — synchronised hyperactivity]
  mech --> auto[Autoimmune — antibody attack]
  deg --> sx[symptoms]
  psy --> sx
  dev --> sx
  vasc --> sx
  par --> sx
  auto --> sx

Skeleton sub-claims

1. Degenerative diseases reveal which substrate each capacity rides on. Alzheimer's targets medial temporal lobe → episodic memory fails. Parkinson's targets substantia nigra → action selection fails. Localisation by attrition.
2. Psychiatric disorders are mostly precision/balance failures, not structural lesions. Schizophrenia, depression, OCD, anxiety: dials, not breaks. Many are trainable — slowly.
3. Developmental disorders show that wiring is the ontology. ADHD, autism, dyslexia: not "deficits" but atypical priors with both costs and benefits. The brain still works; it works differently.
4. Vascular events are time-critical structural lesions. Stroke is plumbing failure; the brain has minutes-to-hours to be saved. Recovery is plasticity-bounded.
5. Paroxysmal disorders (epilepsy, migraine) reveal the cost of stability. The healthy brain sits between hyper- and hypo-synchrony; epilepsy is the failure mode in one direction.
6. Autoimmune brain disease is recent and important. NMDA-receptor encephalitis (Dalmau 2007) showed that what looks like sudden psychosis can be antibodies — a category-changing discovery that's still rewriting the differential.

L2 — Deep dive

degenerative diseases — localisation by attrition

When neurons die slowly, you get to watch the brain disassemble in slow motion. Each major neurodegenerative disease targets a specific protein-aggregation pathology in a specific region:

• Alzheimer's disease — amyloid-β plaques and tau tangles, beginning in entorhinal cortex and hippocampus, spreading laterally. Episodic memory fails first because the binding apparatus is lost; semantic memory and procedural memory lag because they're stored in distributed cortex and basal ganglia. The Braak staging (Braak & Braak 1991) maps the spread. Late-stage AD is global cortical atrophy. What it teaches: episodic memory is hippocampal-dependent; semantic and procedural are not.
• Parkinson's disease — α-synuclein (Lewy bodies) in substantia nigra pars compacta; dopaminergic neurons die; the nigrostriatal pathway fails; the basal ganglia loop loses its gain control. Symptoms: bradykinesia, rigidity, resting tremor, then later cognitive decline. What it teaches: action selection requires dopamine-modulated gating; loss of gain doesn't lose motor programs but loses the ability to initiate them. Hence "kinesia paradoxica" — Parkinson's patients can sometimes run from danger when they cannot walk to the kitchen.
• Huntington's disease — CAG repeat expansion in huntingtin gene; striatal medium spiny neurons die; chorea (involuntary movement) plus cognitive and psychiatric symptoms. Genetically deterministic with anticipation. What it teaches: striatal output normally inhibits unwanted movement; loss of inhibition releases it.
• ALS / motor neuron disease — upper and lower motor neuron loss; progressive paralysis with preserved cognition until late. What it teaches: motor execution is anatomically separable from cognition.
• Frontotemporal dementia (FTD) — tau or TDP-43 in frontal/temporal lobes; behavioural variant presents as personality change before memory loss; semantic variant presents as language loss with preserved episodic memory. What it teaches: personality and social behaviour are frontally-mediated; AD doesn't take them early because AD doesn't start frontally.
• Lewy body dementia — Lewy bodies in cortex; visual hallucinations + parkinsonism + fluctuating cognition. The visual hallucinations are striking — they implicate the same precision-prior mechanism as schizophrenia (see below) but from a different cause.

The general lesson: degenerative diseases are reverse-engineering experiments on a million-subject scale. Each tells us where in the brain a specific capacity lives, by removing it.

psychiatric disorders — precision and balance, not lesion

Psychiatric disorders mostly do not show focal lesions. They show:

• Network-level dysconnectivity — DMN over-active in depression, salience network hyperactive in anxiety, frontal-striatal loops over-coupled in OCD.
• Neurotransmitter imbalances — dopamine (schizophrenia, addiction), serotonin (depression but the simple "serotonin hypothesis" is dead — see Moncrieff et al. 2022), GABA (anxiety), glutamate (depression rapid-acting via ketamine).
• Predictive-coding mis-precision — see humans-as-generators:
• Schizophrenia — prior precision warped (Adams, Stephan, Friston 2013): too-strong priors in some pathways produce hallucinations; too-weak in others produce delusions of meaning. Roughly equal genetic and environmental contribution. Onset typically late teens / early 20s when frontal myelination matures and gates loosen.
• Depression — usually framed as DMN hyperactivity (rumination), reduced reward sensitivity (anhedonia), and elevated cortisol (HPA dysregulation). The "chemical imbalance" narrative is oversimplified; structural changes (hippocampal volume reduction, prefrontal-amygdala coupling) are real and partly reversible with treatment.
• Bipolar disorder — same dial in both directions: depressive low and manic high. Manic episodes look like simultaneous high stack-churn + high prior precision in goal-directed domains — generation maxed out, sleep abandoned, recall preserved acutely but consolidation poor. Lithium remains the most reliable mood stabiliser; mechanism partly understood (GSK-3 inhibition, circadian phase resetting).
• OCD — cortico-striato-thalamo-cortical loop hyperactivity (Saxena & Rauch 2000). The obsession is the stack-monopolisation pattern from humans-as-generators; the compulsion is the only action that briefly relieves the loop. SSRIs help; ERP (exposure
  • response prevention) helps; deep brain stimulation of the anterior limb of internal capsule is reserved for severe refractory cases.
• Anxiety disorders — amygdala-prefrontal imbalance (under-regulation by mPFC); easily chronified by avoidance learning. Most-treatable psychiatric category at the population level (CBT effect sizes ≥ 0.8 in meta-analyses).
• PTSD — trauma installs a high-precision threat prior plus a permission to bypass top-down regulation. Hippocampal volume reduction in chronic cases; whether cause or consequence is contested (Gilbertson twin studies suggest at least partial pre-existing vulnerability).

The unifying frame: psychiatric disorders are mostly dial settings of the same machine that produces normal experience. This is good news for trainability and bad news for "find the lesion." None of the major psychiatric disorders has a single biomarker; all are diagnosed clinically.

developmental disorders — wiring as ontology

Developmental disorders aren't "broken" brains but differently-wired brains, often with both costs and benefits:

• Autism spectrum — atypical sensory gating, weaker top-down priors (Pellicano & Burr 2012: "hypopriors" hypothesis), unusual local-vs-global integration. Many domains hyper-acute (pattern, detail, systemising); social inference taxed because social inference is the most prior-heavy perception. Heterogeneous: spectrum is the right word.
• ADHD — frontal-striatal dopaminergic dysregulation; reward delay impaired; attention variance higher (which sometimes manifests as hyperfocus, not just deficit). Stimulants help by raising tonic dopamine in PFC, paradoxically calming. Useful priors break down at low arousal.
• Dyslexia — phonological processing (left temporo-parietal) atypicality; visual-word-form area connectivity differences. The brain is not trying to read like a typical reader and failing; it's reading via a different pathway. Compensatory strategies and phonics-based training work.
• Tourette syndrome — basal-ganglia disinhibition; tics are released motor patterns. ~50% comorbidity with ADHD/OCD; same loop family.

The lesson: wiring sets the priors. Developmental disorders don't break the generator; they ship a non-default conditioning distribution. Many associated traits are dial-trade-offs (autism's detail-orientation, ADHD's high exploration) that are advantages in some niches.

vascular events — minutes matter

The brain is a high-flow, low-reserve organ. ~15-20% of cardiac output goes to the brain; without flow, neurons die in minutes. Two main vascular failure modes:

• Ischaemic stroke — blood vessel blocked (thrombosis or embolism). The penumbra (tissue that's hypoperfused but not yet dead) can be saved if perfusion is restored within 4.5 hours (thrombolysis) or up to 24h for some large-vessel occlusions (mechanical thrombectomy). "Time is brain": ~1.9 million neurons die per minute of untreated large-vessel occlusion (Saver 2006).
• Haemorrhagic stroke — vessel rupture; blood is itself toxic to neurons. Outcome depends on size and location.

What stroke teaches:

• Localisation is reliable for core deficits. Broca's aphasia (left frontal), Wernicke's aphasia (left posterior temporal), neglect (right parietal), homonymous hemianopia (occipital), hemiparesis (motor strip or internal capsule). The classic neurology textbook syndromes are real because the vasculature is stereotyped.
• Recovery is partly real plasticity, partly de-oedema. The first 1-3 months show steepest recovery; gains continue more slowly for years with practice. Constraint-induced movement therapy (Taub) was the proof that adult plasticity is robust if forced.

paroxysmal disorders — the price of being on the edge

Healthy cortex is critical in the physics sense — close to the edge of synchrony, which gives it dynamic range. Paroxysmal disorders are failures of that balance:

• Epilepsy — synchronised hyperactivity; either focal (partial seizure) or generalised. Causes range from genetic ion channel mutations to scar tissue from old strokes/trauma. ~70% of patients are controlled by anticonvulsants; ~30% are refractory; ~10% are surgically curable. Temporal lobe epilepsy occasionally produces ecstatic auras (Dostoevsky's seizures), implicating insular and limbic circuits.
• Migraine — cortical spreading depression: a slow wave of depolarisation followed by silencing, propagating across cortex at ~3 mm/min. The aura tracks the wave. Partly genetic (CACNA1A in familial hemiplegic). Triptans and CGRP antagonists are the modern pharmacology; many trigger-avoidance regimens are weakly evidenced.

What paroxysmal disorders teach: the brain's normal operating regime is a narrow band. The same mechanism that makes cortex computationally capable (criticality, recurrent excitation) is the one that fails into seizure when its inhibitory restraint is reduced.

autoimmune disease — recent and reframing

Until ~2007, sudden psychiatric or neurological onset in an otherwise-healthy young person was treated as a primary psychiatric event. Then anti-NMDA-receptor encephalitis was characterised (Dalmau et al. 2007): patients (often young women, sometimes with ovarian teratoma) develop psychiatric symptoms, then seizures, then autonomic instability, then coma. Antibodies attack the NMDA receptor; the syndrome is usually reversible with immunotherapy + tumour removal.

This category-changing discovery has expanded:

• Anti-LGI1 encephalitis (faciobrachial dystonic seizures + memory loss).
• Anti-GAD, anti-CASPR2, anti-DPPX, etc. — a rapidly growing list.
• Some "schizophrenia-like" presentations turn out to be autoimmune; the rate is debated.

Lesson: modern medicine discovers new categories. The DSM is a frozen snapshot of mid-20th-century clinical phenomenology; the underlying biology is still being mapped.

what each disease class teaches about the dials

Mapping back to the generator framework:

disease class	broken dial
Alzheimer's	stack capacity (working memory) + recall (consolidation)
Parkinson's	stack-to-action gain (initiation)
Schizophrenia	prior precision (mis-weighted)
Depression	stack diversity (DMN-locked) + reward gain
OCD	stack diversity (one item monopolises)
Bipolar (manic)	stack churn (too high) + sleep
ADHD	prior precision under low arousal
Stroke	structural — depends on territory
Epilepsy	synchrony band (out of healthy range)

Two warnings:

1. Real diseases break multiple dials. The table above is cartoon-resolution.
2. Dials are not independent. Sleep affects all of them; HPA-axis stress affects all of them; the body axis (next page) sets the floor for many of them.

---

sources

• Braak, H. & Braak, E. (1991). Neuropathological stageing of Alzheimer-related changes.
• Saver, J. (2006). Time is brain — quantified.
• Saxena, S. & Rauch, S. (2000). Functional neuroimaging and the neuroanatomy of obsessive-compulsive disorder.
• Adams, R., Stephan, K., Brown, H., Frith, C., Friston, K. (2013). The computational anatomy of psychosis.
• Pellicano, E. & Burr, D. (2012). When the world becomes 'too real': a Bayesian explanation of autistic perception.
• Moncrieff, J. et al. (2022). The serotonin theory of depression: a systematic umbrella review.
• Dalmau, J. et al. (2007). Paraneoplastic anti-N-methyl-D-aspartate receptor encephalitis.
• Taub, E. et al. (1993). Constraint-induced movement therapy.
• Gilbertson, M. et al. (2002). Smaller hippocampal volume predicts pathologic vulnerability to psychological trauma.
• Kandel, E. (2018). The Disordered Mind.