Amyotrophic lateral sclerosis destroys motor neurons — the cells that carry the command to move from the brain to the muscle. It takes both halves of that relay: the upper motor neurons in the motor cortex and the lower motor neurons in the brainstem nuclei and the anterior horn of the spinal cord [1]. What follows is progressive weakness and wasting, usually beginning in one place — a hand, a foot, the voice — and then spreading to other body regions until the muscles of breathing fail. Respiratory failure typically limits survival to two to five years from onset [1], with median survival across studies reported between 20 and 48 months and a minority — 10 to 20 percent — living longer than ten years [2] [3].
It is uncommon. A systematic review of population-based studies put the median incidence in Europe at 2.08 per 100,000 per year (IQR 1.47–2.43) and median prevalence at 5.40 per 100,000 (IQR 4.06–7.89) [4], with worldwide incidence varying by region [5]. The gap between those two numbers is itself informative: for most chronic diseases prevalence vastly exceeds incidence, because people live with them for decades. Here prevalence is only about 2.6 times incidence, which is a direct arithmetic consequence of how short survival is.
About 90 percent of cases are sporadic, with no affected family member; the remaining 10 percent show a family history suggesting autosomal dominant inheritance [1]. That ratio has shaped the field's history — and it has also misled it, because the genes found in the familial 10 percent turned out to explain a fraction of the sporadic majority too.
One common description of ALS is wrong, and worth correcting at the start. It is often said that ALS spares the mind. It does not, entirely. In a study of 279 patients with sporadic ALS, 51 percent showed cognitive impairment on non-motor, non-speed-dependent testing, against 5 percent of controls; cluster analysis divided them into 49 percent intact, 32 percent mildly impaired, 13 percent moderately and 6 percent severely, and 15 percent met criteria for frontotemporal dementia [6]. The clinical review reaches the same place from a different direction: extra-motor manifestations in up to half of patients, and 10–15 percent meeting FTD criteria [1]. Cognition is relatively spared in most patients, and that matters enormously for how ALS is lived — but "the mind is untouched" is not what the data say, and telling families otherwise sets them up badly.
Start here: one protein, two diseases
The single most consequential finding in ALS pathology came in 2006. Ubiquitin-positive, tau- and α-synuclein-negative inclusions had been recognised as the hallmark of both frontotemporal lobar degeneration and ALS, but the protein inside them was unknown. It turned out to be TDP-43 — hyperphosphorylated, ubiquitinated, cleaved into C-terminal fragments, and recovered only from affected regions including hippocampus, neocortex and spinal cord [7] [8].
That identified a common pathological substrate linking two diseases that neurology had treated as separate: one of movement, one of behaviour and language. ALS and frontotemporal dementia are now understood as ends of a spectrum, with formal revised diagnostic criteria for the ALS–FTD spectrum [9] [10], converging molecular mechanisms in RNA and protein homeostasis [11] [12], and TARDBP mutations found in both familial and sporadic ALS shortly after the pathology was identified [13] [14]. TDP-43's normal job is RNA handling, and losing it depletes long pre-mRNAs and causes missplicing — a loss-of-function route to neuronal vulnerability that runs alongside the toxicity of the aggregates themselves [15]. TDP-43 pathology also appears in a distinct late-life amnestic syndrome, LATE, which is not ALS but shares the protein [16].
The genes
The first ALS gene was SOD1, in 1993 [17], and expressing mutant human SOD1 in mice produced motor neuron degeneration — the model on which most of the field's preclinical work has been built [18]. SOD1 was a puzzle: the protein is an antioxidant enzyme, but the disease is not caused by losing that activity. Transgenic mice expressing an altered dismutase established that the disorder "does not result from a diminution of activity and, as such, represents a dominant 'gain of function' mutation" [19], and later work showed motor neuron death arises from a mutant-mediated toxic property, with both elimination and elevation of wild-type SOD1 leaving it unchanged [20]. Work in chimeric and conditional mice showed that onset and progression are determined by different cell types — mutant SOD1 in motor neurons drives onset, while its presence in microglia and astrocytes drives how fast the disease then progresses [21] [22] [23]. ALS is not a disease of motor neurons alone but of motor neurons and their non-neuronal neighbours [24], and it begins distally, as an axonopathy, before the cell body dies [25].
Then in 2011 two groups simultaneously reported that an expanded GGGGCC hexanucleotide repeat in a noncoding region of C9orf72 causes chromosome 9p-linked ALS–FTD [26] [27]. It is the commonest genetic cause of both diseases, produces a distinctive clinical and pathological picture [28], and — being a noncoding repeat that generates toxic RNA foci and RAN-translated proteins — suggested its own therapeutic route [29] [30].
FUS, another RNA-processing protein, followed in 2009 [31] [32], as did mutations in prion-like domains of hnRNPA2B1 and hnRNPA1 [33] and intermediate-length polyglutamine expansions in ataxin-2 as a risk factor [34]. The recurring theme is RNA-binding proteins with aggregation-prone low-complexity domains [12] [35].
How much does this explain? A meta-analysis of 111 studies pooling mutation frequencies for the four major genes found C9orf72, SOD1, TARDBP and FUS together in 47.7 percent of familial ALS and 5.2 percent of sporadic ALS, with significant differences between European and Asian populations [36]. Read that carefully. Half of familial cases still have no identified gene among the big four — and the 5.2 percent in sporadic disease, applied to the 90 percent of patients who are sporadic, means a meaningful number of people with no family history nonetheless carry a causative variant. That is the entire premise of offering genetic testing outside familial cases, and it is why the precision therapy below is not a rare-disease footnote.
Pillar 1: measurement and diagnosis
There is no test for ALS
The diagnosis is clinical. It requires evidence of both upper and lower motor neuron dysfunction, evidence of progressive spread within a region or to other regions, and the absence of another explanation — the El Escorial criteria as originally formulated by the World Federation of Neurology [37] and then revised [38]. Electrodiagnostic criteria set out how EMG evidence of lower motor neuron loss substitutes for clinical signs, which increases sensitivity earlier in the illness [39].
The structural cause of diagnostic delay is written into that definition. A criterion set requiring progression over time and spread across body regions cannot, even in principle, be satisfied at the first visit. Add the need to exclude the ALS mimics — cervical spondylotic myelopathy, multifocal motor neuropathy, Kennedy's disease [40] — and months pass by construction, not by negligence. Whether that delay itself predicts outcome is a separate question, and the evidence is genuinely conflicting [2].
An honest note on scope: this substrate was assembled by citation crawl during an OpenAlex outage and did not return the Gold Coast criteria, the more recent simplification intended to reduce exactly this delay, nor any study quantifying the delay in months. The review therefore does not state a figure for it. The problem is real and well recognised; the numbers were not recoverable here.
Sorting out what ALS is not, and what it also is
ALS is phenotypically heterogeneous even once diagnosed [41] [42]. Regional variants — flail arm and flail leg — have their own natural histories [43], and primary lateral sclerosis, an upper-motor-neuron-predominant condition, progresses far more slowly [44]. Distinguishing these matters for prognosis and for trial eligibility. Cognitive and behavioural assessment belongs in the workup too, given the FTD overlap [9] [10] [6].
Genetic testing
Testing for C9orf72 and SOD1 in particular has moved from academic interest to clinical necessity, because for one of them there is now a drug [36] [35].
Neurofilament: the biomarker that arrived
Neurofilament light chain is released when axons degenerate, and it is raised in ALS in both serum and cerebrospinal fluid [45] [46] [47]. It performs as a diagnostic marker [48], phosphorylated neurofilament heavy subunit is associated with faster ALSFRS-R decline [49], and CSF neurofilament light tracks progression rate [50] [51].
Its most striking use is presymptomatic. In a longitudinal study of 84 at-risk carriers of ALS-associated mutations, 34 controls, 17 patients and — critically — 10 phenoconverters observed both before and after clinical disease emerged, serum and CSF neurofilament light were substantially higher in patients than in controls and at-risk individuals, and remained relatively stable over time in each group [52]. A biomarker that rises before symptoms, in people identified by genotype, is the enabling condition for treating ALS before it starts — and that is exactly what has been designed [53].
Centerpiece: a simple simulatable model of the decline slope
The ALS Functional Rating Scale-Revised runs from 48 down to 0, with higher scores indicating better function [54] [55]; it scores everyday abilities such as speech, swallowing, handwriting, walking and breathing. Its defining property is that in most patients it falls close to linearly over time. The disease is, to first approximation, a straight line down, and the slope of that line is both the strongest available prognostic marker and the primary endpoint of nearly every trial. The rate of symptom progression is an independent predictor of survival [2].
The three trials, and what they each reported.
CENTAUR randomised 137 people with definite ALS and symptom onset within the previous 18 months, and reported the primary outcome as a rate: −1.24 points per month on sodium phenylbutyrate–taurursodiol against −1.66 points per month on placebo, a difference of 0.42 points per month (95% CI 0.03–0.81, P = 0.03) [54].
The edaravone confirmatory study reported a total: change in ALSFRS-R over its 24-week treatment period of −6.35 ± 0.84 on placebo and −5.70 ± 0.85 on edaravone, a difference of 0.65 ± 0.78, P = 0.411 — efficacy not demonstrated [56].
VALOR, the tofersen trial, also reported a total: change from baseline to week 28 in the faster-progression subgroup of −8.14 on placebo and −6.98 on tofersen, a difference of 1.2 points (95% CI −3.2 to 5.5, P = 0.97) [55].
The check the arithmetic was never given. Two of those trials report a total change over a fixed window and never convert it to a rate; the third reports a rate and never reports a total. Dividing each total by its own duration — an operation that is only valid if the decline is approximately linear — gives 1.147 points per month for edaravone's placebo arm and 1.260 for VALOR's. Those agree to within 10 percent, from trials eight years apart, on two continents, with different drugs, different durations and different endpoint definitions. Nothing was fitted. If the underlying decline were not close to linear, the conversion would have no reason to land.
CENTAUR's placebo arm is faster than both, at 1.66. That difference is real and worth naming rather than averaging away: CENTAUR required definite ALS with symptom onset inside the previous 18 months, which selects for people who reached a definite diagnosis quickly — that is, for faster progressors. That reading is an inference from the entry criteria, flagged as such on the figure, not something the trial states.
What a slower slope buys. Taking CENTAUR's two reported rates at face value and extending them, the score reaches half of full function at 14.5 months on placebo and 19.4 months on treatment — a gain of 4.9 months. The figure draws that extrapolation dashed, because it runs well past the 24 weeks either arm was actually observed for, and a linear model fitted over six months is not entitled to predict twenty. It is a translation of the trial result into units patients care about, not a claim the trial made.
The teaching point is why the slope is the endpoint. Because decline is near-linear, a treatment effect shows up as a change in gradient rather than a change in state, and gradients need either large samples or long follow-up to separate. A drug that slows decline by a quarter looks, at three months, almost exactly like a drug that does nothing. That is the structural reason ALS trials are hard, and it sits underneath the field's record: over 60 molecules investigated since riluzole, and the overwhelming majority failing to demonstrate efficacy [3].
Three honest limits. Linearity is an approximation — real trajectories bend, particularly near the floor of the scale and around events like starting ventilation, and the linear fit is best over the middle of the range. The ALSFRS-R is a sum of ordinal items, so a point lost in speech and a point lost in walking count equally, which they do not to the person losing them. And all three slopes here come from trial populations, which are selected for eligibility and are not the general ALS population.
Pillar 2: treatment
The honest summary: there is no cure, the disease-modifying drugs are of modest effect, and the interventions that most change survival and quality of life are supportive rather than pharmacological.
Riluzole
Riluzole modulates glutamatergic transmission, on the hypothesis that excitotoxicity contributes to motor neuron death. The pivotal trial randomised 155 outpatients and found 12-month survival of 74 percent with riluzole against 58 percent with placebo (P = 0.014), with a striking effect in bulbar-onset disease (73 versus 35 percent) and a smaller but still significant overall advantage by the end of the placebo-controlled period [57]. Deterioration of muscle strength was also slower.
The Cochrane assessment is the number to quote to a patient: riluzole "probably prolongs survival ... by about two months", the beneficial effects are "very modest", and the drug is expensive [58] [59] [60]. Two months is not nothing in a disease measured in a few years. It is also not what the 74-versus-58 headline sounds like, and the difference between those two framings is the difference between informed consent and disappointment.
Edaravone
Edaravone is a free-radical scavenger, and its evidence is genuinely mixed. The confirmatory study described above failed: the reduction in ALSFRS-R was smaller with edaravone but the difference was not significant (P = 0.411) [56]. A subsequent trial restricted to a well-defined subgroup of patients — earlier, faster-progressing, better respiratory function — was positive and led to approval [61]. Reviews have treated riluzole and edaravone together as a study in how modest ALS efficacy gets established [62] [63], and the route from a failed endpoint to regulatory approval has itself been examined [64].
The fair reading is that edaravone works in a narrow population defined partly by the failure that preceded it, and that whether it helps patients outside that window is not established.
Tofersen: the precision milestone, and an instructive failure
Tofersen is an antisense oligonucleotide that reduces synthesis of SOD1 protein, delivered intrathecally, for people with SOD1-associated ALS. The approach was demonstrated first in the SOD1 mouse and then in a first-in-human intrathecal study [65], followed by a phase 1–2 trial [66].
Then VALOR. Tofersen missed its primary endpoint outright — the 28-week ALSFRS-R difference was 1.2 points with a P value of 0.97 [55]. But it did what it was designed to do at the molecular level: tofersen produced greater reductions in CSF SOD1 concentration and in plasma neurofilament light than placebo. The target engaged; the function, over 28 weeks, did not follow. In the open-label extension at 52 weeks, the change in ALSFRS-R was −6.0 in participants who started tofersen at entry against −9.5 in those who switched from placebo at week 28, a difference of 3.5 points (95% CI 0.4–6.7) [55].
That is the whole argument in miniature. A drug that unambiguously hits its molecular target, misses on function at six months, and shows a separation at a year — which is exactly the pattern the slope model predicts for a real but partial effect on a near-linear decline. It is also why the field moved to the obvious next question: give it before the decline starts. A phase 3 trial was designed to initiate tofersen in clinically presymptomatic SOD1 carriers identified by neurofilament elevation [53] [52].
What actually extends life
Non-invasive ventilation. A randomised trial of NIV in ALS examined effects on survival and quality of life [67], and the systematic review of prognostic factors concludes plainly that non-invasive positive-pressure ventilation "has been found to improve survival" [2]. Among everything in this section, respiratory support has the best claim to changing the natural history.
Nutrition is less settled: the same review states that the effect of enteral nutrition on survival "is still unclear" [2], even though nutritional status is itself related to outcome. Gastrostomy sits within the nutritional management the practice parameters set out [68] [69]; the evidence that it extends life is weaker than most people assume.
Multidisciplinary care — neurology, respiratory, speech and language, dietetics, physiotherapy, occupational therapy, palliative care — is the framework the practice parameters are built around [68] [69]. The parameters also cover the parts of care that no drug touches: symptom management, communication aids as speech is lost, and advance-care planning, which in a disease with this trajectory needs to happen early enough that the patient can still participate in it.
Pillar 3: what is unresolved
Gene-targeted therapy beyond SOD1. The C9orf72 repeat expansion generates toxic RNA foci and RAN-translated dipeptide proteins, and antisense intervention mitigates that toxicity in models by targeting sense and antisense foci [70] [71] [29] [30]. Translating that into people is the field's next major test, and C9orf72 is far commoner than SOD1.
Treating before symptoms. Neurofilament rises before phenoconversion in at-risk carriers [52], and a trial has been designed on that premise [53]. If it works, it changes what ALS is clinically — from a disease diagnosed after irreversible loss to one intercepted during it.
Why trials fail. Over 60 molecules since riluzole, and a near-uniform record of failure [3], including agents that looked convincing in the SOD1 mouse. Lithium was reported to delay progression in 2008 [72] and did not become a treatment; the general record is the context in which any single such report should now be read [3]. Recent trials continue in the same pattern [73] [74]. Part of the problem is the slope endpoint's low resolution over short windows, part is population heterogeneity, and part is that the standard mouse model represents a mutation carried by a small minority of patients. The systematic review of prognostic factors makes concrete recommendations on this: randomise on age, respiratory status and a pre-entry measure of progression rate, and consider natural-history controls and futility designs [2].
What ALS actually is. Sporadic ALS remains largely unexplained — a "conspiracy of genes, environment and time" [75], with roughly half of even familial cases outside the four major genes [36]. TDP-43 is the near-universal pathology [7], but whether the damage is dominated by loss of its normal RNA function [15] or by the toxicity of what it forms is not settled [11] [12].
Where the disease starts. Onset and progression are separably controlled, in different cell types [21], and the earliest failure appears to be distal, at the axon [25] — a century after Charcot, the question of why these particular neurons die remains open [76].
Dig deeper in lmmol
ALS completes the neurodegenerative set, and reading it against the others is where the pattern shows. Parkinson's disease is the closest instructive contrast: it also has a long presymptomatic phase and a protein aggregate at its centre, but it has effective symptomatic therapy and ALS has essentially none. Huntington's disease shares the antisense-oligonucleotide strategy and the same idea of treating genetically identified people before onset [53] — and shares with C9orf72 the biology of a repeat expansion [26]. Alzheimer's disease shares TDP-43 more directly than is usually appreciated, since the LATE consensus describes a TDP-43 encephalopathy that clinically mimics it [16]. Multiple sclerosis is the sharpest contrast of all: another disease of the central nervous system tracked by a functional scale and a neurofilament biomarker [47] [77], but where disease-modifying therapy works — the difference being that MS damage is immune-mediated and interruptible, while ALS is degenerative and, so far, is not. The full collection is at health.