Cells and synapses

In typical development, the number of synapses in the human cortex peaks in infancy or early childhood and then falls, a process called pruning. Net loss runs through late childhood and into mid-adolescence,¹ and in the prefrontal cortex spine loss continues into the third decade of life.²

Three small studies of brain tissue donated after death report more dendritic spines in autistic people than in non-autistic people: in frontal, temporal and parietal cortex,³ in the temporal lobe,⁴ and in the amygdala, in children and adolescents only.⁵ Each region rests on a single study. In the first, the highest spine densities went with lower brain weight and were most common in autistic people with lower cognitive scores.³

The most cited of these, by Tang and colleagues in 2014, counted spines in 10 autistic and 10 non-autistic brains aged about 2 to 20; the adolescent comparison was in males only. Spine density was slightly higher in the autistic children and clearly higher in the autistic adolescents. Between the childhood and adolescent groups, spine density fell by about 41 to 45% in the non-autistic tissue and by about 16% in the autistic tissue.⁴ The amygdala study, with usable data from 14 autistic and 13 non-autistic brains (from donors aged 7 to 46), found higher spine density under 18 and no difference in adults.⁵

Scans of living adults give a mixed picture. In one small sample, a PET study of 12 autistic adults (6 of them men) found synaptic density about 17% lower across the cortex than in 20 non-autistic adults, and lower density went with more autistic traits.⁶

mTOR is a protein that acts as a hub inside cells: it takes in signals about the energy available and drives cell growth and the building of new proteins.⁸ The genes TSC1, TSC2 and PTEN normally hold it in check. Harmful changes in them cause tuberous sclerosis complex (TSC) and PTEN hamartoma tumor syndrome, in which mTOR is overactive.^8,9

Autism is common in both. One meta-analysis found autistic traits in about 36% of people with TSC,¹⁰ and another found them in about 25% of people with a PTEN mutation, an estimate its authors say should be read cautiously.¹¹ Among the children one clinic referred for PTEN testing, a selected group, 5 of 60 autistic children carried a mutation, and every carrier had a head size more than 2 standard deviations above average; the clinic's cohort was defined partly by large heads, so this link partly reflects who was tested.¹² High rates of autism in these syndromes do not by themselves show that mTOR causes the autism, and blocking mTOR with everolimus did not change autistic traits in trials (see below).^13,14

Outside these syndromes, findings point in different directions. In temporal-lobe tissue from 13 autistic and 16 comparison cases, Tang and colleagues found higher mTOR activity in the autistic adolescents (13 to 20), not in the younger children.⁴ A study of the fusiform gyrus, in 11 autistic and 13 comparison brains, all autism without a known cause, found lower levels of several proteins in the same pathway. Its authors suggest the pathway may be disturbed in either direction.¹⁵ Two studies of blood cells, from 41 and 33 autistic children, mostly boys aged 3 to 11, report higher activity in the pathway.^16,17 They come from separate groups.

In young mice carrying one faulty copy of the Tsc2 gene, mTOR was overactive, autophagy was blocked and fewer spines were pruned. Rapamycin, a drug that blocks mTOR, given from about 3 to 4 weeks of age restored pruning and social behavior. It did not do so in mice whose neurons could not carry out autophagy, and the authors concluded that autophagy in neurons is needed for this pruning.⁴

Separately, microglia engulf synaptic material and are needed for normal pruning in mice.¹⁸ Mice lacking autophagy in a family of immune cells that includes microglia had more spines and differences in social and repetitive behavior.¹⁹ These are findings about autophagy in two different kinds of cell, and neither study shows mTOR stopping microglia from pruning.^4,19 The only human data on this chain are correlational: in Tang's post-mortem tissue, more spines went with overactive mTOR and lower autophagy markers.⁴

Rapamycin has also reversed some changes in adult mice. In adult mice with one faulty copy of Tsc2, a brief course restored synaptic plasticity and learning.²⁰ A 2026 study used mice whose mothers were given a small dose of a bacterial toxin early in pregnancy to cause inflammation. In the adult offspring, a single dose of rapamycin reduced over-excitable neurons, seizure susceptibility, repetitive grooming and circling, and sensory over-responsiveness within 2 hours, and supplementary data show social interaction improving too.²¹ The effect was short-lived: repetitive behavior returned within 72 hours, and daily doses given for 5 weeks stopped working. The study did not measure dendritic spines or autophagy.²¹ Autism-like behavior in these mice means scores on animal tests, not traits a person would recognize.

Everolimus, a drug in the same family as rapamycin, has been tested for thinking and behavior in people with TSC.²² In a US trial of 47 children and young people aged 6 to 21, six months of everolimus did not improve neurocognition or behavior on nearly all measures.¹³ In a Dutch trial of 32 children aged 4 to 17 with a low IQ, a learning disability or autism (60 had been planned), twelve months of everolimus did not improve IQ or autistic traits.¹⁴ A Cochrane review found that everolimus shrinks TSC-related tumors and reduces seizures, though more people on it needed a lower dose or stopped; for thinking and behavior it found only one eligible trial, of 42 people, which showed no difference (low certainty).²² In a single trial of 46 people aged 5 to 45 with PTEN hamartoma tumor syndrome, the main outcome did not improve, while some secondary measures shifted slightly toward improvement, by small amounts that were not statistically significant.⁹

In 2003, Rubenstein and Merzenich proposed that some forms of autism involve a raised ratio of excitation to inhibition in brain systems for sensation, memory, social behavior and emotion.²³ A 2019 update by one of the original authors restates the idea in terms of reduced signal-to-noise in circuits, refines what the balance means, and proposes it as an organizing framework. Its evidence comes largely from mice carrying autism-linked gene changes.²⁴

A 2022 review argues that the idea is hard to test. Researchers rarely say at which level they mean the balance (chemical messengers, single cells, local circuits or the whole brain), and there are no validated ways to measure it in people. It argues that using newer EEG markers of the balance in clinical groups is premature until their link to the underlying biology is understood.²⁵

A raised ratio may also be the brain adjusting rather than the source of a difference. In four mouse models of autism, including Tsc2, the ratio of excitation to inhibition in somatosensory cortex was raised, mainly through weaker inhibition, but firing was largely normal; the authors read the shift as a compensating adjustment that keeps activity stable.²⁶ A 2015 review argues that in many cases it is hard to separate an initial change in excitation or inhibition from the brain's compensation for it.²⁷

Magnetic resonance spectroscopy (MRS) can estimate levels of GABA and glutamate, the brain's main inhibitory and excitatory chemical messengers, in living people.²⁸ Meta-analyses report lower GABA on average in autistic people.^29,30,31 The earliest rested on 5 small studies with about 150 participants in total, autistic and comparison combined.²⁹ In a 2024 meta-analysis the difference was significant in children but not in adults, and results varied with how studies were run. Few adult studies existed, and the authors caution that the age difference may reflect this.³⁰ A 2023 meta-analysis also found higher glutamate in the prefrontal cortex and lower NAA, a chemical used as a marker of healthy neurons.^31,30

Blood is a separate and more indirect index than brain tissue. A 2026 meta-analysis of blood studies, in about 800 autistic and 720 comparison participants, found higher glutamate and lower glutamine on average, but no clear difference in GABA, with very high heterogeneity between studies and very low certainty.⁴⁴

Studies of autistic adults, each from a different group and each measuring different regions, disagree. One of 25 adults with an IQ of 80 or more found no change in GABA in the striatum or medial prefrontal cortex, and lower glutamate in the striatum.³² One of 29 young adults found no differences in visual, auditory or motor cortex, with statistics that favored no difference.²⁸ One of 43 adults found higher GABA in the left dorsolateral prefrontal cortex.³³

EEG and MEG offer another indirect measure: the slope of background brain activity, which some researchers read as a sign of the balance. In 49 boys aged 6 to 15, the slope was flatter only in autistic boys with an IQ below 85.³⁴ A study of 63 autistic adolescents and adults found a flatter slope;³⁵ a study of 35 autistic adults found at most a weak overall difference, with the autistic adults' slope steepening during an active task.³⁶ The studies used different methods and conditions.

Bumetanide, a diuretic, lowers chloride inside neurons, which is meant to strengthen inhibition through GABA.³⁸ Two early trials by one French research group, in 60 and 88 children, reported improvement, partly in analyses of subgroups or of those who completed treatment.^38,39 A group in Shanghai reported benefit in a placebo-controlled trial of 119 children aged 3 to 6, mostly with moderate to severe autism, on the Childhood Autism Rating Scale, and its authors say the clinical significance remains uncertain.⁴⁰ The same group reported benefit in a trial of 83 young children in which the comparison group received neither bumetanide nor a placebo. The two trials come from one group and count as one line of evidence.⁴¹

An independent trial of 92 children aged 7 to 15 in the Netherlands found no effect on its main outcome; one of three secondary measures, repetitive behavior, improved, a finding not yet repeated, and low potassium occurred in 51% of children on bumetanide.⁴² Two larger trials run by one company, of 211 children each (aged 7 to 17 and 2 to 6), were stopped early because bumetanide did no better than placebo on any outcome, and the company ended its development for autism. The senior author of the 2012 French trial is the senior author of the report on these trials.^43,38

Two meta-analyses published in 2026 pool these trials. Both report a small short-term improvement on the Childhood Autism Rating Scale in children and adolescents, significant at three months but not at six, with high heterogeneity between studies.^45,46 One rates the certainty of the evidence low to very low;⁴⁶ the other reports improvement on a clinician-rated global impression but no significant benefit on the Social Responsiveness Scale-2 and inconsistent results on repetitive behavior depending on the tool, and finds the improvement on the rating scale below the 4.5-point change its authors treat as clinically important.⁴⁵

Synaptic pruning is a cellular idea that autistic writers often use. Devon Price describes typical development as "a process of slowly pruning unhelpful connections", and writes that in autistic brains some regions stay hyperconnected throughout life while others are relatively underconnected; Price cites researchers who describe autistic development and pruning as "disrupted".^c1

Sol Smith builds a whole explanation of autism on pruning. Smith writes that autistic people have underactive mirror neurons, and proposes that a brain out of step with the people around it cannot tell which synapses to prune; Smith says plainly that no study has drawn this conclusion.^c2 In a later chapter Smith writes that while other children trim unused pathways to match the people around them, "we tend to just leave a bunch there, resulting in a Seussian train yard with interchanges that aren't immediately efficient."^c3 Smith uses the same idea to explain why some autistic people use their native language "more like a second language", choosing words for their literal meaning.^c4

Smith also writes that autistic brains have more synaptic connections, "11 percent more is the number usually found in postmortem brain comparisons".^c5 Nick Walker, an autistic scholar of the neurodiversity paradigm, writes that the central difference is that autistic brains are "characterized by particularly high levels of synaptic connectivity and responsiveness", so that each piece of information lands harder and less predictably.^c6 The research above is less settled. Three small post-mortem studies, each of a different brain region, report more dendritic spines, one of them only before adulthood, and scans of living adults found fewer synapses in one small study and no difference in another. The often-said box above covers the claim that autistic brains do not prune. The link from mirror neurons to pruning is Smith's own hypothesis; the research sections above do not cover mirror neurons.