Key findings
- 80%the approximate heritability of ASD; the proposed burden is framed as acting on the timing and severity of presentation, not on liability itself
- 17 of 20of the adolescents in Ghaziuddin's case series of late regression showed catatonia; onset across the series averaged around age 13
- 3:1 to 1.2:1the male-to-female diagnosis ratio in Swedish registry data, near 3:1 through age 10 and roughly 1.2:1 by age 20
- 6testable predictions the author sets out, any of which would count against the model if the results were negative
Abstract
Autism spectrum disorder (ASD) research has treated early regression, late regression, late diagnosis without documented childhood symptoms, and the sex disparity in diagnostic timing as separate phenomena. We propose that early regression, late regression, and genuinely late-emerging late diagnoses may share one process in some cases: cumulative oxidative and inflammatory burden from environmental chemical exposure, infection and autoimmune activity (including PANS/PANDAS, maternal immune activation, and folate receptor alpha autoantibody-mediated cerebral folate deficiency), and chronic psychosocial stress including adverse childhood experiences, mediated in part by gut-microbiome-dependent folate and B12 status. ASD liability is largely genetic; the burden is proposed to act on the timing and severity of presentation, not as an alternative. Individual, including sex-linked, buffering capacity determines whether and when the burden crosses a functional threshold. Developmental timing of any such crossing, rather than the specific exposure, determines presentation: no phenotype, early-onset ASD, early or late regression, or, in genuinely late-emerging cases, gradual late diagnosis. Belated recognition of early-present traits is outside the model. A speculative buffering hypothesis complements, and does not replace, the dominant masking account of the sex disparity. This is a theoretical synthesis rather than an established causal model: no study yet measures these factors together, and we outline the prospective evidence needed to test it. The model does not, at present, support any specific treatment for ASD or regression.
Introduction
Autism research has long kept a set of related developmental phenomena in separate silos. Early regression, the loss of previously acquired skills, typically in toddlerhood, is its own well-established subfield. A rarer and more recently documented pattern, sometimes called "late regression," describes symptom drift emerging only after a period of apparent developmental stability. Some people come to meet criteria in adulthood that they did not meet in childhood (a “gained diagnosis”). And a substantial, disproportionately female population is not diagnosed until adolescence or adulthood despite no documented loss of childhood skills at all. Each of these four patterns tends to be studied with its own methodology, within its own age range, by researchers who rarely cross into one another’s literature.
This paper’s starting hypothesis is that these four patterns may share a common root: a cumulative burden of oxidative stress, built up through the joint action of environmental exposure and psychosocial stress, and channeled in part through disruption of the gut microbiome and the one-carbon metabolic pathway it supports. It is bounded by genetic liability (Section 3.5) and by the distinction between genuine change and belated recognition (Section 2.4). We present this explicitly as a hypothesis-generating synthesis, not an established causal account: no existing study has measured oxidative-stress markers, microbiome composition, B-vitamin status, cortisol, and exposure history together in a single cohort spanning early regression, late regression, and late diagnosis. Our contribution here is not new data but a proposed mechanistic thread linking these literatures, together with a roadmap for the evidence that would be needed to test it.1–3
Section 2 defines the phenomena; Sections 3-6 review the exposure, immune, microbiome and stress evidence; Section 7 addresses sex differences; Section 8 presents the integrated model and its testable predictions; Section 9 covers limitations.
Phenomenology: Defining the Forms of "Regression"
In everyday and clinical usage alike, "regression" is asked to cover several distinct phenomena at once. This section pulls them apart, since the evidence base and plausible mechanism differ substantially from one to the next. It also includes two later-diagnosis patterns (Sections 2.3 and 2.4) so that all four can be compared.
2.1 Early regression
The best-documented form of regression involves the loss of skills a child had already acquired, most commonly expressive language, though social engagement, eye contact, and motor skills can be affected too, following what had been roughly a year of typical development.4 The bulk of documented cases occur between 12 and 36 months, clustering around 21–24 months of age, though one large epidemiological study extended the plausible window considerably, from as early as 6 months out to age 7.2,5,6
How common regression turns out to be depends heavily on how it is measured. Parent-report studies looking backward typically find it in roughly a third of ASD children, a figure a meta-analysis pooled at close to 32%.7 Medical and school records tell a more conservative story, near 21%, though this likely undercounts cases that depend on a parent noticing the change and a clinician writing it down. Studies that follow children forward in time rather than asking parents to recall the past tell yet another story: one found social-skill decline somewhere between 6 months and 3 years of age in 86% of children eventually diagnosed with ASD, hinting that regression may be a much more common, and more gradual, phenomenon than the classic image of sudden loss suggests.2,8
What actually causes early regression remains an open question. One account, from Ozonoff and colleagues, holds that the apparent "loss" may not be a discrete neurological event at all but a gradual fading of skills, as an underlying drive toward social communication goes insufficiently reinforced.8 Seizure and epileptiform activity have been examined as possible drivers but have not held up consistently across larger studies.9 The MMR-vaccine hypothesis deserves mention only to be dismissed: the originating study was retracted by The Lancet in 201010 and the hypothesis has since been discredited by large epidemiological studies and meta-analyses.11,12
2.2 Late regression after a period of stability
A much smaller literature addresses the opposite pattern: decline that sets in only after a period of relative stability, implying a child who had, up to that point, been doing comparatively well. The main systematic account here is Ghaziuddin’s case series, which describes marked functional decline after stable childhood functioning in 20 adolescents with an established autism diagnosis, with onset averaging around age 13. The earliest signs were often obsessive slowing and ritualistic behavior, followed by mood and behavioral change; catatonia showed up in 17 of the 20 cases, and outcomes in this cohort were frequently poor. This remains a small, case-series-level literature rather than a population-based one, but it is enough to establish adolescent-onset functional decline as a real, if under-studied, clinical entity.13 A more recent, larger sample likewise concerns catatonia rather than regression in general: among 300 autistic adolescents and young adults, catatonic presentation tracked closely with stressful life events. However, an association of this kind cannot show that stress precipitates it.14
Catatonia in autism is a recognized syndrome with specific and often effective treatments, notably benzodiazepines and electroconvulsive therapy13, so it should be treated as a distinct, treatable state and not folded into regression in autism generally. This has a direct clinical consequence: an adolescent decline should not be attributed to a cumulative-burden process while a treatable syndrome (catatonia, depression, emerging psychosis, or medical illness) remains possible, and the model proposed here is relevant, if at all, only after such conditions have been considered and, where present, treated.
2.3 The "gained diagnosis" trajectory
A related but more rigorously documented phenomenon is what Elias and Lord (2022) call "Gained ASD Diagnosis." Tracking 155 individuals from age 2 to 25 with blinded assessments at each wave, the authors sorted participants into four trajectories, Retained ASD, Lost ASD, Never Had ASD, and Gained ASD Diagnosis, the last describing people whose adult diagnosis exceeded whatever they had been classified as in childhood. Novel diagnoses emerged mostly in adulthood regardless of cognitive level; among participants with typical-range IQ, the underlying pattern was a gradual worsening of symptom severity scores rather than a sudden relapse, sometimes not clearing the diagnostic bar until adulthood.15
Elias and Lord read this as evidence that autism classification genuinely shifts across development, rooted in gradual changes to core features starting in childhood rather than simple diagnostic error, at least within their cohort. It is a single study, and no mechanism for rising severity has been established.15 The pattern is not confined to clinical samples: a prospective UK birth cohort followed to age 25 turned up a distinct "late-emerging" autistic-trait trajectory in a meaningful slice of the general population, and a later review of symptom-severity change across the lifespan likewise concludes that rising, not just falling, severity is a recognized, if less common, pattern.16,17
2.4 Late diagnosis without documented childhood symptoms
A conceptually separate, though practically overlapping, phenomenon is late diagnosis in people with no documented childhood symptoms whatsoever, which could reflect genuine late emergence or simply belated recognition of traits that were there all along. Even the definition is unsettled: a 2025 systematic review spanning 420 studies found that barely a third bothered to specify what counted as "late" at all, and among those that did, cutoffs averaged 11.5 years but ranged anywhere from age 2 to 55, clustering bimodally around age 3 and age 18.18
Large cohorts make clear just how common late diagnosis really is: in the Simons Foundation Powering Autism Research for Knowledge (SPARK) cohort, roughly half of autistic adults weren’t diagnosed until after age 17, and many not until their 30s, 40s, or 50s.19 A Danish registry study is especially telling: among adults diagnosed as autistic, 69% of men and 61% of women had no record of any related psychiatric or neurological diagnosis before age 18, a pattern the study’s authors say is hard to explain as simple misdiagnosis or diagnostic overshadowing. The Millennium Cohort Study points the same direction: late diagnosis there was predicted by typical-range intelligence and an absence of early parental concern, the profile of a child managing well enough until adolescent social and academic demands catch up with them.20,21 A related Millennium Cohort report found that late-diagnosed children accumulate greater social and emotional difficulty, which fits a cumulative-threshold model but equally fits the consequences of delayed recognition and support.22
Late regression occurs in people who already carry an autism diagnosis; in gained and late diagnosis, someone newly meets criteria, through genuine emergence, gradual accumulation, or, probably often, a blend with delayed recognition that cannot currently be teased apart.
The late-diagnosis group should itself be treated as at least two populations, and the distinction determines what the model is asked to explain. The first is the late-recognized: people whose traits were present in early development but went unnoticed or were attributed to other causes. This is the group the diagnostic criteria anticipate, since Criterion C of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) requires that symptoms be present in the early developmental period while allowing that they may not become fully manifest until social demands exceed limited capacities.23 The second is the genuinely late-emerging: people in whom symptoms appear to arise or intensify after the early developmental period, as in the late-emerging trait trajectory in the UK birth cohort and the “gained diagnosis” trajectory described above. Only this second group falls within the scope of the framework.
Figure 1 lays out these four patterns side by side, ordered by the developmental timing that sets each one apart.
Oxidative Stress and Environmental Contributors
3.1 Rationale: oxidative stress as a developmental vulnerability
Oxidative stress, an imbalance between reactive oxygen species production and antioxidant capacity, has been repeatedly documented in ASD populations, including reduced glutathione levels and altered markers of oxidative damage relative to neurotypical peers.24–26 The developing brain is thought to be particularly vulnerable to oxidative insult given its high metabolic demand, relatively high lipid content, and comparatively immature antioxidant defense systems in early life. This vulnerability is not confined to infancy: puberty and adolescence involve substantial neurodevelopmental remodeling and correspondingly elevated metabolic demand, offering a plausible second window of heightened sensitivity, one that aligns temporally with the late-regression pattern described in Section 2.2 (mean onset around age 13).1,2
A pattern recurs across the specific exposure classes reviewed in Section 3.2, direct particle neurotoxicity, endocrine disruption, air pollution, tire-derived chemicals, heavy metals and pesticides, and ultraprocessed food intake: each is reported to converge on a small, overlapping set of mechanisms, oxidative stress and mitochondrial dysfunction, blood–brain and placental barrier disruption, neuroinflammation and impaired microglial synaptic pruning, endocrine and nuclear-receptor disruption, and epigenetic modification of ASD-related genes. Primary studies support individual links in this list, including impaired corticogenesis27,28, disrupted microglial synaptic pruning29, oxidative stress and ferroptosis30,31, placental transfer and nuclear-receptor disruption32, endocrine and epigenetic disruption33, and inflammatory or metal-related effects in air-pollution and flame-retardant cohorts34–36. The primary studies cited here cover micro- and nanoplastics, plastic-associated and tire-derived chemicals, and air pollution including airborne metals; for pesticides, other heavy-metal exposures, and ultraprocessed food, the convergence claim currently rests on the review literature. Rather than functioning as independent, competing explanations, these converging mechanisms plausibly represent multiple entry points into a single shared final pathway: disruption of the tightly time-constrained developmental sequence of neurogenesis, corticogenesis, synaptogenesis, and activity-dependent synaptic pruning.1 Pollutant class, timing, and dose determine which point in the sequence is disrupted, while the convergent phenotype reflects failure of the same developmental program.
Individual susceptibility to this oxidative burden is not uniform, and genetic variation in folate-pathway and mitochondrial genes offers one explanation for why comparable exposure produces regression, or ASD more broadly, in one child and not another. The broader genetic architecture of ASD is addressed in Section 3.5.
Folate transport into the central nervous system depends primarily on folate receptor alpha (FRα) and the reduced folate carrier (RFC); mutations in FOLR1 impair FRα function directly, while the SLC19A1 polymorphism rs1051266 reduces RFC efficiency, and either route can independently produce cerebral folate deficiency regardless of dietary folate intake, with associated decreases in social communication and emotional recognition. Methylenetetrahydrofolate reductase (MTHFR) polymorphisms37, which reduce the conversion of folate to its active 5-methyltetrahydrofolate form, represent a third, partially overlapping route to the same functional endpoint of reduced one-carbon metabolic capacity.2 The SLC19A1 rs1051266 variant has been examined directly in relation to ASD and cerebral folate deficiency38, and folate receptor autoantibodies provide a separate, autoimmune route to cerebral folate deficiency39.
A separate genetic vulnerability, involving mitochondrial function rather than folate transport, bears more directly on the regression phenotype addressed throughout this paper: the sequence of dysfunction differs by regression status, with inflammation and immune dysregulation appearing first in non-regressed children, but mitochondrial dysfunction appearing first in children who go on to regress, with any effect on immune activity emerging only later.2,40,41 This suggests mitochondrial vulnerability may be a distinct, earlier-acting susceptibility factor for the regression subtype, one that oxidative burden (Section 3.2) could also stress directly. Cord blood metabolomics offers one emerging, non-invasive biomarker candidate for this pathway: machine-learning analysis of cord blood metabolic profiles has proven more predictive of later ASD diagnosis than maternal mid-gestational blood in both male and female offspring,42 and elevated cord blood acylcarnitine specifically predicts both ASD and attention-deficit/hyperactivity disorder (ADHD) outcomes43, though neither finding has yet been tested against the environmental exposures reviewed here.2
3.1.1 Emotional and Psychosocial Stress. Emotional and psychosocial stress belongs alongside the chemical, dietary, and infectious/autoimmune contributors addressed elsewhere in this section as a distinct route to oxidative burden. Maternal stress during pregnancy plausibly disrupts fetal brain development directly, through elevated cortisol and pro-inflammatory signaling, and chronic stress in early childhood may carry a parallel risk for immune and neurological outcomes (for a review of prenatal mechanisms, see 44). The primary evidence is indirect: prenatal stress has been linked to ASD-related outcomes in murine models45,46 and to ASD severity in a cross-sectional study of autistic children47, and prenatal exposure to synthetic glucocorticoids has been associated with later mental disorders in offspring, including ASD (Section 6.1)48. None of these studies measured stress-related cortisol and cytokines together or addressed chronic early-childhood stress, so this mechanism remains a hypothesis. On this model, a family under chemical and psychosocial stress simultaneously accumulates oxidative burden along convergent pathways.1,2
3.1.2 Adverse Childhood Experiences and Prolonged Hypothalamic–Pituitary–Adrenal (HPA) Axis Activation. Adverse childhood experiences (ACEs), sustained abuse, neglect, household dysfunction, or food and housing insecurity during childhood, plausibly constitute a distinct, chronic route to oxidative burden, in contrast to the acute prenatal stress pathway of Section 3.1.1. Cumulative ACE burden tracks cumulative psychiatric risk in children and adolescents,49 consistent with the dose-response relationship established by the original ACE Study,50 although neither is autism-specific. Trauma-related DNA methylation signatures predicted adult psychiatric and other adverse outcomes better than reported trauma history,51 suggesting a plausible epigenetic route by which chronic early-life stress could produce durable changes relevant to neurodevelopment, analogous to the one-carbon methylation pathways of Section 5. Because the exposure is sustained over years, ACE-driven burden more closely resembles the cumulative-threshold model (Section 3.4) than a single insult; autism-specific ACE evidence is addressed in Section 6.2.
3.2 Chemical and dietary exposure: a brief synthesis
The pollutant classes most directly relevant to this convergence argument, micro/nanoplastics, plastic-associated endocrine-disrupting chemicals, ambient and indoor air pollution, tire wear particles, heavy metals and pesticides, industrial chemicals, and ultra-processed food intake, are each reviewed in full mechanistic and epidemiological detail in a companion paper,52 which grades the evidence class-by-class and should be consulted for the primary literature. Two findings bear specifically on the phenomena addressed here. First, a mouse model using polyethylene microplastics directly tested and supported a microplastic–ASD hypothesis: prenatal exposure produced impaired social interaction and repetitive behavior, altered brain glucose metabolism, and ASD-associated gut microbiome shifts, tying this exposure route directly to the microbiome pathway developed in Section 5.53 Second, in the Childhood Autism Risks from Genetics and the Environment (CHARGE) cohort, parental occupational exposure to plastics and polymers was linked to greater irritability, hyperactivity, and weaker adaptive skills in ASD children, a rare human severity, rather than diagnosis-risk, correlate.54 Across the full pollutant literature, late gestation and the second postnatal year recur as sensitive windows,34,55,56 and male sex recurs as a vulnerability modifier,33,57,58 both themes relevant to the developmental-timing argument of Section 8.2. All findings are observational (Section 9).
3.3 Infections and autoimmune-triggered regression
Infection- and immune-triggered regression is the most directly relevant existing evidence for a stable baseline followed by sudden decline. Pediatric Acute-onset Neuropsychiatric Syndrome (PANS) and its streptococcus-specific subtype, Pediatric Autoimmune Neuropsychiatric Disorders Associated with Streptococcus (PANDAS), describe a pattern in which infection triggers a misdirected immune response in a genetically susceptible child, producing brain inflammation and an abrupt shift in neuropsychiatric symptoms.59 The defining feature of PANS/PANDAS, a sudden shift from a relatively stable baseline, is the abrupt form that late regression (Section 2.2) can take, and contrasts with the more gradual symptom drift described for the "gained diagnosis" trajectory (Section 2.3). PANS/PANDAS should not be read as a purely childhood phenomenon, however. The original PANDAS criteria were framed around prepubertal onset, but the PANS criteria set no upper age limit, and post-pubertal presentations have been described59. The mechanism relevant to this paper is therefore better framed broadly as post-infectious autoimmune neuroinflammation, of which PANS/PANDAS are specific, recognized forms, so that it plainly extends to the adolescent window in which late regression occurs (mean onset about 13 years; Section 2.2). The model does not require infection-triggered late regression to be abrupt: it allows abrupt, subacute, or mixed onset (the Ghaziuddin cases began with obsessive slowing, a more insidious course), and no study has established which tempo characterizes an infection-triggered subset. The overlap of PANS/PANDAS with autism remains poorly quantified, and no population-based study has isolated PANS/PANDAS rates within an autistic cohort.60 The closest primary data concern autistic regression more broadly: a clinical study identified features suggestive of autoimmunity and neuroinflammation in patients with autistic regression61, but it did not address PANS/PANDAS. Separately, folate receptor alpha autoantibodies have been examined in a PANS/PANDAS population62, which links the syndrome to the folate pathway (Section 4) without quantifying its overlap with autism.
Recent large-cohort and clinical-consensus work has sharpened both the epidemiology and the mechanism behind this pattern. A 2023 retrospective review of over 90,000 pediatric patients across three primary care populations estimated the annual incidence of PANDAS/PANS at approximately 1 in 11,765 children ages 3–12,60 an age band that excludes adolescence, so it does not show that these syndromes are confined to childhood, and the American Academy of Pediatrics' first clinical report specifically addressing PANS, published in 2025, now provides consensus diagnostic and management guidance. Supporting, though observational, evidence for an infection-driven mechanism comes from a 2025 study reporting that pediatric influenza infection itself, rather than antiviral treatment, was associated with an increase in neuropsychiatric events, predominantly neurologic, and oseltamivir was associated with a smaller increase; this is suggestive rather than conclusive because the study did not address PANS, psychiatric illness, or regression.63 A parallel mechanistic account, still preprint but directly relevant to this paper's oxidative-and-inflammatory-burden framework, proposes that maternal immune activation (MIA) imprints a lasting regulatory T-cell deficiency in offspring, removing a key brake on neuroinflammation and creating a durable susceptibility to autism-like phenotypes that could plausibly also lower the threshold at which a later infection precipitates PANS/PANDAS-type regression.64
This same MIA/autoimmune literature also converges directly with the environmental-pollutant burden reviewed in Section 3.2: maternal infection and inflammation, environmental toxicant co-exposure, maternal autoimmune disease, and folate receptor alpha autoantibody (FRAA)-mediated cerebral folate deficiency (CFD) have been proposed as four distinct but mechanistically overlapping routes into the same downstream cascade of chronic neuroinflammation, oxidative stress, mitochondrial dysfunction, and disrupted synaptic pruning.3 PANS/PANDAS is thus one recognizable instance of a general principle: a postnatal infection precipitating regression in a child whose neuroimmune buffering is already reduced. Primary studies bear on individual arms and not on the convergence itself: maternal immune activation and later ASD or socio-adaptive outcomes65–67, maternal autoantibodies68,69, toxicant exposure and maternal immune responses35,36, folate receptor autoimmunity in cerebral folate deficiency39,70, and air-pollution exposure and folate status in pregnancy71. No study cited here shows that the four routes converge on a single cascade, so that convergence is inferred from the separate findings.
Down syndrome regression disorder offers a useful and potentially testable comparison for the infection- and immune-triggered arm of the model. It is an adolescent or young-adult onset regression with catatonic features, proposed immune contributions, and reported responses to immunotherapy, which together make it a clinical setting in which an immune-triggered regression mechanism can be examined in a population with a known genetic substrate.72 Pre-onset burden markers could be compared between those who later regress and those who do not (Section 8.5).
3.4 Convergent, multi-hit exposure logic
Our central methodological claim is that no single exposure category, chemical, dietary, infectious, or psychosocial, is likely to be sufficient on its own to explain regression or late diagnosis in most individual cases. Rather, the model proposed here treats these exposures as additive or synergistic contributors to a cumulative oxidative burden, such that regression or symptom emergence occurs when that burden crosses an individual threshold, one that is presumably shaped by genetic antioxidant capacity, baseline exposure history and, as discussed in Section 7, possibly sex.
This convergence logic follows the author's earlier review of the MIA/autoimmune/environmental literature.3 A large systematic review and meta-analysis of cohort studies similarly finds that environmental pollutant exposures function as risk modifiers rather than independent sufficient causes, consistent with a threshold model in which cumulative exposure across multiple pathways, rather than any single exposure category, determines whether regression or symptom emergence occurs.73
This convergence framing carries an important epistemic caveat that we acknowledge explicitly: a recent large population-based analysis examining maternal immune activation and autism causation directly addressed the question of whether the MIA-ASD association reflects a genuine causal pathway or is substantially attributable to familial confounding, genetic and environmental factors shared between a mother's health status and her child's neurodevelopmental outcome, rather than direct in-utero exposure effects.74 If the MIA-ASD association is substantially explained by familial confounding, then MIA is not a causal input to oxidative burden, and that arm of the model is weakened rather than absorbed into it. The MIA component of the model (Sections 3.3 and 4) is therefore treated here as provisional, supported only to the extent that designs able to account for shared familial factors show a causal contribution. More generally, the convergent-burden model is plausible and mechanistically grounded but untested against confounding across all of its component pathways.
Figure 2 summarizes how these seven exposure sub-mechanisms, converge on the shared final pathway described above, and how genetic liability and folate-pathway modifiers relate to it.
3.5 Scope of the model: genetic architecture and what the burden is claimed to explain
The framework in this paper should not be read as an environmental account of ASD liability. Twin and population studies place the heritability of ASD at roughly 80%,75 and rare de novo and inherited variants of large effect are a major part of its architecture. In Figure 2, the folate-pathway genes shown are those relevant to the one-carbon mechanism and are not a summary of ASD genetics; they are treated as modifiers of susceptibility to the burden, not principal determinants of liability.
Accordingly, the burden proposed here is framed in two bounded ways. First, it is proposed to act on residual variance: among individuals with a given level of inherited liability, environmental and psychosocial burden may help determine whether, and when, a diagnostic threshold is reached. Second, and more specifically, it is proposed as a determinant of the timing and severity of presentation, including regression, and not of liability itself. The model has most to say about presentations in which inherited liability is moderate and timing is variable; early-onset ASD without regression may be largely genetically determined.
Maternal Immune Activation, Autoimmune Disease, and Cerebral Folate Deficiency
This section addresses a family of immune mechanisms distinct from, but mechanistically related to, the infection-triggered PANS/PANDAS pathway discussed in Section 3.3. Where PANS/PANDAS describes a postnatal, infection-triggered immune event producing abrupt regression, the mechanisms addressed here, maternal immune activation, maternal autoimmune disease with autoantibody transfer, and cerebral folate deficiency via folate receptor alpha autoantibodies, operate substantially in the prenatal period and carry independent evidentiary support for a role in ASD risk and symptom severity generally. This section treats them as a parallel, more thoroughly characterized route to the same convergent oxidative and inflammatory burden our model depends on.3 The causal status of the MIA component is contested, however (Section 3.4), and this section describes plausible mechanisms rather than established causal inputs.
Each of these three mechanisms, maternal immune activation via pro-inflammatory cytokine signaling across the placenta, maternal autoimmune disease with brain-reactive autoantibody transfer, and folate receptor alpha autoantibody-mediated cerebral folate deficiency, is reviewed in full mechanistic and treatment detail in previously published work3; we summarize only what bears on regression, late diagnosis, and this paper’s convergence argument. Primary studies exist for each mechanism. Maternal immune activation during pregnancy has been associated with ASD risk and with more difficulty in socio-adaptive behavior in autistic children65,66. Maternal autoantibody profiles have been examined as biomarkers of ASD69, and exposure to ASD-specific maternal autoantibodies altered behavior and brain structure in a rat model68. Folate receptor autoantibodies have been described in cerebral folate deficiency syndrome39, and a milk-free diet lowered folate receptor autoimmunity in that syndrome70. All three converge on the same downstream targets: microglial activation and impaired synaptic pruning, oxidative and mitochondrial stress, and, for cerebral folate deficiency specifically, a proposed homocysteine-mediated oxidative mechanism. Support for the homocysteine mechanism is indirect and comes from work outside ASD: folate depletion with elevated homocysteine promoted oxidative stress in rat liver76, and folate treatment improved vascular redox state in patients with coronary artery disease77.
4.1 Synthesis: autoimmune mechanisms within this paper's oxidative-stress framework
Microbially derived folate can modulate host folate receptor expression, so gut microbial composition might shift receptor availability independently of autoantibody status. Whether pollutant-driven microbiome disruption (Section 3.2) or family stress (Section 6) increases FRAA production or MIA-triggering infection is untested and a research priority.3,78
Figure 3 summarizes how these three prenatal immune mechanisms converge on shared downstream targets.
The Gut Microbiome as a Mediating Pathway
The gut microbiome plays a documented role in regulating host oxidative stress and in synthesizing and modulating bioavailability of the B vitamins central to the body’s antioxidant and methylation machinery, making it a plausible mediating layer between the environmental exposures reviewed above and downstream neurodevelopmental vulnerability; this evidence, including the microbial folate-synthesis capacity and diet-driven dysbiosis pathways, has been reviewed in detail elsewhere.52
5.1 B9/B12 and methylation capacity
Folate (vitamin B9) and vitamin B12 are essential cofactors in the one-carbon metabolic cycle, which governs methylation reactions throughout the body, including in processes relevant to neurodevelopment and to the synthesis of glutathione, the body's primary intracellular antioxidant. Reduced bioavailability of either vitamin, whether from dietary insufficiency, gut absorption impairment, or microbiome composition that disfavors their synthesis or supports their consumption by competing microbes, would be expected to reduce both methylation and antioxidant capacity simultaneously, providing a direct biochemical link between microbiome disruption and oxidative vulnerability.78,79
5.2 Existing regression-microbiome evidence
At least one study has directly examined the question of a regression-specific microbiome signature: a comparison of fecal microbiota in ASD children, who are with versus without documented mental regression. Evidence specific to regression remains scarce and suggestive rather than confirmatory.80
Another test of this section's core premise, that early gut microbial composition predicts later regression or diagnosis, comes from the Markers of Autism Risk in Babies–Learning Early Signs (MARBLES) prospective cohort (n=137), which analyzed fecal samples collected between 0 and 7 months of age via 16S rRNA sequencing against clinical classification at around 36 months. The study found no significant differences in alpha diversity, beta diversity, or any single differentially abundant bacterial taxon between infants who went on to develop ASD and those who remained typically developing. The cohort was small and enriched for elevated familial likelihood of ASD, the authors themselves note that the study may have been underpowered, and it reported uncorrected trends in individual taxa, so the overall picture is better described as mixed than as a definitive null. This tempers, without ruling out, the mechanistic case built from the bacterial folate-synthesis and diet-microbiome literature above; earlier effects, in utero or via the maternal microbiome, would predate this sampling window.81
A further problem concerns the direction of causality. Yap and colleagues found that most associations between ASD and gut microbiome composition were explained by the restricted dietary preferences common in autistic children,82 which suggests that microbiome differences may be a consequence of autistic feeding patterns and not a cause. If food selectivity itself shapes the microbiome, case-control differences cannot be taken as evidence for that link, and the regression-specific comparison described above is open to the same objection. The microbiome pathway is therefore retained here as a hypothesis about a possible mediating route whose direction of effect has yet to be shown in prospectively sampled, diet-adjusted data.
5.3 Proposed causal chain
Synthesizing Sections 3, 4, and 5, we propose the following working causal chain: environmental exposure (chemical, dietary, or infectious/autoimmune) disrupts gut microbiome composition; microbiome disruption reduces bioavailability of B9/B12 and related cofactors; reduced cofactor availability impairs both methylation and glutathione-dependent antioxidant capacity; reduced antioxidant capacity lowers the threshold at which a given exposure, or a subsequent infection, or a period of psychosocial stress (Section 6), precipitates visible regression or symptom emergence. Most individual links in this chain have some independent evidentiary support; the chain as a whole has not been tested. The first link is further uncertain because of the diet-related reverse-causation problem described in Section 5.2, so the chain is best read as conditional on that link.
Figure 4 summarizes this proposed causal chain and the genetic factors that modify specific links in it. The first link, from exposure to microbiome disruption, is marked as the least certain.
Psychosocial Stress and Cortisol
6.1 The HPA axis and oxidative stress: a bidirectional relationship
Cortisol, the HPA axis’s principal output, sits in a two-way relationship with oxidative stress: sustained HPA activation raises markers of oxidative damage, and oxidative stress in turn can throw HPA reactivity off balance, potentially locking the two into a self-sustaining loop under prolonged stress. That two-way street matters for the model here, since it means psychosocial stress doesn’t have to act alone to raise regression risk, it can also magnify whatever oxidative damage the environmental exposures in Section 3 are already doing.1
The most substantial large-scale human evidence for the glucocorticoid arm of this relationship (the study did not measure oxidative stress) comes from a Danish cohort of over a million infants, comparing children exposed in utero to systemic glucocorticoids, prescribed for either preterm-delivery risk or maternal autoimmune/inflammatory disease, against unexposed children whose mothers shared the same underlying condition. Exposed children carried a higher adjusted 15-year ASD risk in both groups (RR 1.5 and 1.3, respectively), along with elevated ADHD and mood/anxiety/stress-disorder risk, a pattern that held up under sibling-matched sensitivity analysis.48 The authors point to fetal HPA-axis programming and altered CNS development following excess glucocorticoid exposure at specific gestational windows as the likely driver, the same pathway this paper invokes for endogenous, stress-driven cortisol. Exogenous glucocorticoids and chronic maternal stress acting through endogenous cortisol are not the same exposure, so the study supports the maternal-stress pathway only by analogy.
6.2 Family and psychosocial stress evidence in ASD
Research on psychosocial stress specifically in autistic populations is still developing but active. Kern and colleagues built the Childhood Adversity and Social Stress Questionnaire (CASS-Q) to measure adverse and traumatic experiences and trauma symptoms in autistic youth; its symptom subscale, notably, lists skill regression and increased self-injury as trauma-related outcomes in their own right, treating regression as something stress can produce rather than purely a neurodevelopmental event. More broadly, autistic children are documented to face ACEs at higher rates than their non-autistic peers, especially in lower-income households, and autistic adults report more adverse events across their lifespan than neurotypical adults do. Animal work lends this some biological plausibility: early-life stress increases susceptibility to autoimmune neuroinflammatory disease and worsens its course, a finding that ties directly back to the infection/autoimmune pathway covered in Section 3.3.83–88
The direction of causality in the ACE literature also needs to be stated plainly. Higher ACE rates in autistic children may partly reflect consequences of autism, such as bullying, family strain, and the poverty associated with caregiving, and not causes of symptom change. Cross-sectional associations between adversity and autistic outcomes, including those obtained with the CASS-Q, cannot separate the two directions. The stress arm of the model, here and in Section 3.1.2, is accordingly offered as a hypothesis whose direction of effect is unresolved.
6.3 Proposed synergy with environmental exposure
Family and psychosocial stress, in this model, function as an additive, and possibly synergistic, oxidative load stacked on top of chemical and infectious exposure rather than a rival explanation for regression or late symptom onset. Given the two-way HPA–oxidative relationship laid out in Section 6.1, a child already carrying oxidative burden from chemical exposure, diet, or microbiome disruption (Sections 3–5) may be more susceptible to psychosocial stress tipping them into visible regression than a child without that baseline load.
6.4 Caveat
It is worth being direct about a gap here: no study has directly linked ACE burden to regression episodes in already-diagnosed autistic children the way the PANS/PANDAS literature links infection to regression. No study has tested stress burden as a predictor of regression onset; this is an open research priority. A second gap is reverse causation, as described in Section 6.2.
Figure 5 lays out this two-way HPA–oxidative-stress relationship and how it’s proposed to add to the environmental burden covered elsewhere in this paper.
Sex Disparity in Diagnostic Timing, Reframed
7.1 The data
A clear picture of the sex disparity in diagnostic timing emerges from recent large-sample data. Epic Research’s 2024 analysis found the median diagnosis age for boys had fallen to about 5 (from 7 in 2015), while girls’ median held nearly flat around 8. Close to a quarter of girls were diagnosed only in adulthood, versus 12% of boys.89
Swedish registry data tracking 2.76 million people born 1985–2020 found a male-to-female diagnosis ratio near 3:1 through age 10, declining steadily thereafter; diagnoses peaked at 10–14 for males but not until 15–19 for females, and the cumulative ratio had dropped to roughly 1.2:1 by age 20 as of 2022. CDC ADDM surveillance shows a similar, if milder, narrowing in childhood prevalence ratios, from 4.2:1 in 2018 to 3.4:1 in 2022 among 8-year-olds.90,91
7.2 The dominant explanation: masking and diagnostic criteria
The best-supported explanation remains that autism presents differently, on average, in girls and women than the male-typical profile diagnostic criteria and clinician training were built around, including a greater capacity for social "masking" and special interests that read as less overtly atypical. Meta-analytic evidence suggests that the true male-to-female ratio is nearer 3:1 than the 4:1 often quoted, and that girls who meet criteria are less likely to be diagnosed,92 and a study of self-reported camouflaging reports gender differences in how autistic adults mask their traits.93 Nothing in this paper displaces that explanation.
7.3 A complementary hypothesis: sex differences in oxidative buffering
We offer a complementary, not competing, hypothesis: sex differences in oxidative-stress buffering, estrogen’s antioxidant properties, say, or differences in HPA-axis/cortisol reactivity, may help explain why female presentation tends to run subtler or later, independent of anything to do with clinician masking. Under the threshold model built up in Sections 3–6, if female physiology offers somewhat more buffering against a given cumulative burden, that burden would on average take longer to cross a diagnostic threshold, which is in the direction of later median diagnosis age and higher share of adult diagnoses seen in girls and women. Related exposure studies show sex-specific effects (prenatal bisphenol A and brain aromatase disruption in male ASD-like mice33; fetal exposure to persistent organic pollutants57; air pollution and folate status71) but do not test buffering itself.
Three difficulties with this hypothesis should be stated. First, male predominance is already present in early childhood, when sex differences in circulating estrogen are minimal outside the neonatal period, so an estrogen-based antioxidant explanation cannot account for early sex differences in timing; if sex-linked buffering matters at that age, it would have to operate through a different mechanism, such as HPA-axis reactivity, which is less well characterized. Second, the rise in female diagnoses and the reported female late presentations occur across puberty, when estrogen rises; on a buffering account one might expect the opposite pattern, with later or fewer threshold crossings after puberty. The hypothesis does not fit pubertal timing in any obvious way, and we flag this as unresolved. Third, the natural home for a sex-specific account of ASD is the female protective effect literature. Siblings of affected girls show more autistic traits than siblings of affected boys,94 and affected females carry a higher burden of rare deleterious variants,95 both of which suggest that females need a greater etiological load to reach the diagnostic threshold. The buffering hypothesis differs in kind (how long a burden takes to cross the threshold, rather than how much liability is needed); the protective-effect literature is far better supported, and the oxidative-buffering idea should be read, at most, as a speculative additional contribution to timing.
7.4 The narrowing ratio as indirect, non-definitive support
The Swedish finding that the male-to-female ratio narrows to near 1:1 by age 20 is consistent with, though it doesn’t prove, a cumulative-threshold account. A pure masking/recognition account predicts the same pattern equally well, though, and existing data can’t tell the two apart. We flag this as a direction for future work (Section 9), not as evidence already banked for the oxidative-buffering hypothesis.91
Figure 6 lays out these two complementary explanations side by side, along with the narrowing-ratio pattern both are consistent with.
Integrated Model
Pulling Sections 3 through 7 together, we propose the following integrated model. Chemical, dietary, and infectious/autoimmune exposures combine with family and psychosocial stress to raise cumulative oxidative and inflammatory burden. Gut microbiome disruption mediates part of that burden by cutting B9/B12 bioavailability, which weakens both methylation and glutathione-dependent antioxidant capacity. This combined burden then runs up against individual buffering capacity, shaped by genetics, prior exposure history, and possibly sex-linked differences in antioxidant and HPA-axis function, and it is that interaction that decides whether, and when, a visible behavioral threshold gets crossed. Inherited genetic liability remains the larger determinant of whether ASD occurs at all (Section 3.5); the model concerns mainly the timing and severity of presentation. The developmental timing and tempo (abrupt or gradual) of any threshold crossing then determine the clinical presentation (Section 8.2).
8.1 Multiple convergent paths to a shared burden
No single exposure class is being proposed as necessary or sufficient on its own. A child could plausibly reach the same functional burden through chemical exposure alone, through chronic psychosocial stress alone, or, more typically in this model, through a combination of moderate exposures spread across several domains, none of which would register as significant in a single-exposure study.
8.2 Critical-period timing and the divergence of clinical presentation
What decides the clinical presentation, in this model, isn’t just how large the cumulative burden gets but when in development it crosses a functional threshold, if it crosses one at all, and whether that crossing happens abruptly or gradually. Five broad outcomes follow. If buffering capacity outpaces lifetime burden, the threshold is never crossed and no ASD phenotype emerges, the implicit baseline against which the other four are defined. If the threshold is crossed prenatally or very early postnatally, typically through heavy chemical or infectious burden during the late-gestation window that Section 3.2 repeatedly flags as vulnerable, the result is early-onset ASD with no regression history, the classic, most-studied presentation. If early burden is substantial but still sub-threshold, a second discrete hit, a PANS/PANDAS-qualifying infection (Section 3.3), an acute household stressor, or further chemical exposure, arriving around 12–24 months could push the total over threshold abruptly, producing the toddler-regression pattern from Section 2.1. If burden stays sub-threshold through early childhood but a later hit lands, chronic ACE exposure (Section 3.1.2), a subsequent infection, or accumulating chemical burden, after a stretch of apparent stability, the result is late regression, whose roughly age-13 mean onset (Section 2.2) is compatible with a chronic ACE/HPA-axis route, though it is equally compatible with the pubertal remodeling window raised in Section 3.1. And if burden simply accumulates gradually with no abrupt crossing, paired with comparatively strong buffering, symptoms may surface slowly enough to resemble the Elias and Lord “gained diagnosis” trajectory from Section 2.3, in which symptom severity rises gradually from an earlier, lower classification until criteria are met, rather than a discrete regression event. Only genuinely late-emerging late diagnosis follows this gradual route; late-recognized diagnosis is not a threshold-crossing event and lies outside the model.
8.3 Compensation, masking, and the sex disparity
Buffering capacity is doing two jobs at once here, and the distinction has clinical teeth. For some children, enough buffering prevents the threshold from ever being crossed, so no ASD phenotype develops no matter how the underlying burden is measured. For others, buffering doesn’t stop the crossing but slows or softens it, yielding a milder, more compensable phenotype that’s harder for caregivers and clinicians to read as autism, especially against diagnostic criteria and training historically built around a male-typical presentation (Section 7.2). The sex-linked buffering hypothesis (Section 7.3) and masking (Section 7.2) are not mutually exclusive, and current data cannot separate their contributions (Section 7.4).
8.4 Schematic model
Figure 7 sketches this model schematically.
8.5 Operational definitions and testable predictions
The constructs used in this model have so far been descriptive. To make them testable, we define them as follows, as working proposals to be refined and not as validated measures.
Oxidative burden is defined as a composite, longitudinally measured index of redox and inflammatory status, combining, for example, the reduced-to-oxidized glutathione ratio, lipid and DNA oxidation products (such as isoprostanes and 8-hydroxy-2'-deoxyguanosine), and inflammatory markers, together with exposure biomarkers where available (for the classes in Section 3.2) and records of infection and psychosocial adversity. The index is an exposure-weighted summary and not a single assay. Components of this kind have been measured in ASD, including transsulfuration and glutathione biomarkers24, markers of cellular stress25, bioenergetic and inflammatory markers40, biomarkers of mitochondrial dysfunction in a systematic review and meta-analysis26, and metabolomic signatures96. Outside ASD, 8-hydroxy-2'-deoxyguanosine has been measured as a marker of DNA damage in adolescents97 and high-sensitivity C-reactive protein as an inflammatory marker in adults98.
Buffering capacity is defined as the ability to restore redox balance after a challenge, indexed by baseline glutathione and B9/B12 status, folate-pathway genotype, and, where feasible, the response of redox markers to a naturally occurring or standardized challenge such as acute illness.
Threshold is treated as an empirical quantity and not an assumed one: the level of the burden index, adjusted for buffering capacity, at which the probability of a defined clinical event (documented loss of skills on standardized assessment, or onset of criterion-level symptoms) rises sharply, estimable as a change-point in prospective data.
With these definitions the model makes predictions that could fail. (1) Pre-onset burden: in children followed prospectively, the burden index measured before onset should be higher, after adjustment for genetic liability, in those who later regress, or whose symptoms later emerge, than in matched peers who do not; if it is not, the model is not supported. (2) Late-recognized versus late-emerging: burden should distinguish genuinely late-emerging individuals from late-recognized individuals whose traits were documented in early childhood; if the two groups do not differ, the distinction drawn in Section 2.4 carries no biological weight and the model's claim for the late-emerging group is undercut. (3) Infection as a second hit: among children with comparable infections, those with higher baseline burden or lower buffering capacity should be more likely to regress afterward; if regression after infection is no more common at higher baseline burden, the multi-hit logic does not hold.
(4) Direction of effect: exposure and burden measured before diet, family circumstances, or diagnosis could have changed them (for example, in antenatal or infant samples) should predict later outcomes; associations present only at or after diagnosis would be consistent with reverse causation, as for the diet-microbiome and ACE arms (Sections 5.2 and 6.2). (5) Genetic liability: if measured genetic liability, polygenic or rare-variant, accounts for the variation in timing among the presentations in Section 2 with no added contribution from burden, the model fails as an account of timing. (6) Sex: the buffering hypothesis predicts that, at equal burden, females reach the threshold later and that redox buffering differs by sex in the relevant age windows; burden measures that do not mediate sex differences in timing once masking is accounted for, or an absence of sex differences in buffering in those windows, would count against it (Section 7.3).
Negative results on any of these points would count against the model as formulated here. Prospective cohorts with biobanked antenatal and early-childhood samples, enriched for later regression, are the most plausible way to test them.
Limitations and Required Future Evidence
This paper offers a theoretical synthesis rather than new empirical data; its contribution is the proposed mechanistic link between literatures, not a demonstrated causal pathway. Testing it directly would require a prospective cohort measuring oxidative-stress markers, gut microbiome composition, B9/B12 status, cortisol, and detailed exposure history simultaneously in children tracked from early childhood into adolescence, with enough power to compare those who go through early regression, late regression, late diagnosis, and typical early diagnosis. No such study yet exists.
Fairness requires naming the alternative explanations that don’t need this model at all. Late diagnosis in particular could largely reflect evolving diagnostic criteria, shifting clinical awareness, and the masking explanation from Section 7.2, with no oxidative-stress contribution whatsoever. The “gained diagnosis” trajectory Elias and Lord identified15 could just as easily reflect changes in measurement or instrument sensitivity across development rather than a genuine biological process. None of these rules out the model proposed here, but a rigorous test would need to show an oxidative/exposure signal on top of what these more established explanations already cover.
Several further limitations apply. First, this is a narrative synthesis and not a systematic review: no systematic search was performed, sources were identified non-systematically, and relevant studies, particularly those reporting contrary or null findings, may have been missed. Second, the argument relies heavily on secondary sources. Much of the mechanistic case in Sections 3 through 6 rests on the author’s own earlier reviews, which summarize the primary studies and do not replace them, and some cited sources are preprints, surveillance summaries, or web reports; an independent systematic review would be needed to confirm that the primary literature supports the claims drawn from it. Third, reverse causation remains possible for the microbiome and ACE arms (Sections 5.2 and 6.2). Fourth, confounding is unmeasured in most of the cited work; familial and genetic confounding in particular may account for exposure-outcome associations such as the MIA-ASD association (Section 3.4). Fifth, the environmental-exposure literature is largely observational and subject to publication bias and selective reporting, so apparent consistency across exposure classes may overstate the strength of the evidence. Sixth, the central constructs were not previously operationalized, which made the model difficult to falsify in practice; Section 8.5 proposes definitions and predictions, but they are untested. Finally, ASD is highly heritable (Section 3.5), the late-diagnosis group is heterogeneous (Section 2.4), and the evidence for late regression is thin and dominated by catatonia (Section 2.2), so the reach of the model is narrower than a general account of why autism emerges when it does.
Implications for treatment. Because this topic attracts antioxidant, folinic acid, antimicrobial, and immunomodulatory treatments of uncertain benefit, it should be stated plainly that the model proposed here does not, at present, support any specific treatment for ASD or regression. It is a hypothesis about risk and timing that has not been tested clinically, and nothing in it should be read as a recommendation to use any of these interventions. Treatable conditions such as catatonia warrant standard evaluation and care regardless of this model.
Conclusion
We have argued that early regression, late regression, and the genuinely late-emerging subset of late diagnoses in autism may, in some cases, share a mechanism, acting on a background of largely genetic liability: cumulative oxidative stress built up through environmental exposure and psychosocial stress, and channeled through the gut microbiome and one-carbon metabolism. The sex-disparity hypothesis is a speculative complement to the masking account (Sections 7.3 and 7.4). This is a research agenda, not a settled conclusion. Should future prospective work bear it out, the public-health implication would be modest: cutting environmental exposure burden and buffering family/psychosocial stress could serve as general protective measures for neurodevelopmental outcomes, regardless of when, or whether, any given person ends up with an autism diagnosis. It does not, at present, support any specific treatment (Section 9).
Declarations
- Conflicts of Interest Statement
- The author has no conflicts of interest to declare.
References
- Ayoub G. Neurodevelopment of Autism: Critical Periods, Stress and Nutrition. Cells. 2024;13(23):1968. doi:10.3390/cells13231968
- Ayoub G. Autism Spectrum Disorder as a Multifactorial Disorder: The Interplay of Genetic Factors and Inflammation. International Journal of Molecular Sciences. 2025;26(13):13. doi:10.3390/ijms26136483
- Ayoub G. Neurodevelopmental Impact of Maternal Immune Activation and Autoimmune Disorders, Environmental Toxicants and Folate Metabolism on Autism Spectrum Disorder. Current Issues in Molecular Biology. 2025;47(9):721. doi:10.3390/cimb47090721
- Furley K, Mehra C, Goin-Kochel RP, et al. Developmental regression in children: Current and future directions. Cortex. 2023;169:5-17. doi:10.1016/j.cortex.2023.09.001
- Goin-Kochel RP, Esler AN, Kanne SM, Hus V. Developmental regression among children with autism spectrum disorder: Onset, duration, and effects on functional outcomes. Research in Autism Spectrum Disorders. 2014;8(7):890-898. doi:10.1016/j.rasd.2014.04.002
- Boterberg S, Charman T, Marschik PB, Bölte S, Roeyers H. Regression in autism spectrum disorder: A critical overview of retrospective findings and recommendations for future research. Neuroscience & Biobehavioral Reviews. 2019;102:24-55. doi:10.1016/j.neubiorev.2019.03.013
- Tan C, Frewer V, Cox G, Williams K, Ure A. Prevalence and Age of Onset of Regression in Children with Autism Spectrum Disorder: A Systematic Review and Meta-analytical Update. Autism Res. 2021;14(3):582-598. doi:10.1002/aur.2463
- Ozonoff S, Iosif AM. Changing conceptualizations of regression: What prospective studies reveal about the onset of autism spectrum disorder. Neurosci Biobehav Rev. 2019;100:296-304. doi:10.1016/j.neubiorev.2019.03.012
- Ruffini R, Brinciotti M, Giovannone F, Pisani F, Tofani M, Sogos C. Correlations between EEG abnormalities and clinical phenotypes in a population of children with autism spectrum disorder. Research in Autism. 2025;123:202536. doi:10.1016/j.reia.2025.202536
- Lancet TE of T. Retraction—Ileal-lymphoid-nodular hyperplasia, non-specific colitis, and pervasive developmental disorder in children. The Lancet. 2010;375(9713):445. doi:10.1016/S0140-6736(10)60175-4
- Taylor LE, Swerdfeger AL, Eslick GD. Vaccines are not associated with autism: an evidence-based meta-analysis of case-control and cohort studies. Vaccine. 2014;32(29):3623-3629. doi:10.1016/j.vaccine.2014.04.085
- Hviid A, Hansen JV, Frisch M, Melbye M. Measles, Mumps, Rubella Vaccination and Autism: A Nationwide Cohort Study. Ann Intern Med. 2019;170(8):513-520. doi:10.7326/M18-2101
- Ghaziuddin M. Catatonia: A Common Cause of Late Regression in Autism. Front Psychiatry. 2021;12:674009. doi:10.3389/fpsyt.2021.674009
- Öksüzoğlu ME, Rollas T, Erdoğan IM, et al. Catatonia in autism spectrum disorder: Analysis of clinical characteristics, stressful life events, and validation of the attenuated behavior questionnaire (ABQ). Psychiatry Res. 2025;351:116643. doi:10.1016/j.psychres.2025.116643
- Elias R, Lord C. Diagnostic stability in individuals with autism spectrum disorder: insights from a longitudinal follow-up study. J Child Psychol Psychiatry. 2022;63(9):973-983. doi:10.1111/jcpp.13551
- Riglin L, Wootton RE, Thapar AK, et al. Variable Emergence of Autism Spectrum Disorder Symptoms From Childhood to Early Adulthood. Am J Psychiatry. 2021;178(8):752-760. doi:10.1176/appi.ajp.2020.20071119
- Waizbard-Bartov E, Miller M. Does the severity of autism symptoms change over time? A review of the evidence, impacts, and gaps in current knowledge. Clin Psychol Rev. 2023;99:102230. doi:10.1016/j.cpr.2022.102230
- Russell AS, McFayden TC, McAllister M, et al. Who, when, where, and why: A systematic review of “late diagnosis” in autism. Autism Res. 2025;18(1):22-36. doi:10.1002/aur.3278
- Cremer S, Antezana L, Conner CM, Eack SM, Mazefsky CA, Northrup JB. The relationship between age of autism diagnosis and life satisfaction in adulthood. Autism. 2026;30(4):931-943. doi:10.1177/13623613261416672
- Rødgaard EM, Jensen K, Miskowiak KW, Mottron L. Childhood diagnoses in individuals identified as autistics in adulthood. Mol Autism. 2021;12(1):73. doi:10.1186/s13229-021-00478-y
- Hosozawa M, Sacker A, Mandy W, Midouhas E, Flouri E, Cable N. Determinants of an autism spectrum disorder diagnosis in childhood and adolescence: Evidence from the UK Millennium Cohort Study. Autism. 2020;24(6):1557-1565. doi:10.1177/1362361320913671
- Mandy W, Midouhas E, Hosozawa M, Cable N, Sacker A, Flouri E. Mental health and social difficulties of late-diagnosed autistic children, across childhood and adolescence. J Child Psychol Psychiatry. 2022;63(11):1405-1414. doi:10.1111/jcpp.13587
- American Psychiatric Association, ed. Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, Text Revision (DSM-5-TR). 2022. Accessed October 11, 2026. https://doi.org/10.1176/appi.books.9780890425787
- Geier DA, Kern JK, Garver CR, Adams JB, Audhya T, Geier MR. A prospective study of transsulfuration biomarkers in autistic disorders. Neurochem Res. 2009;34(2):386-393. doi:10.1007/s11064-008-9782-x
- Dong D, Zielke HR, Yeh D, Yang P. Cellular stress and apoptosis contribute to the pathogenesis of autism spectrum disorder. Autism Res. 2018;11(7):1076-1090. doi:10.1002/aur.1966
- Frye RE, Rincon N, McCarty PJ, Brister D, Scheck AC, Rossignol DA. Biomarkers of mitochondrial dysfunction in autism spectrum disorder: A systematic review and meta-analysis. Neurobiol Dis. 2024;197:106520. doi:10.1016/j.nbd.2024.106520
- Hua T, Kiran S, Li Y, Sang QXA. Microplastics exposure affects neural development of human pluripotent stem cell-derived cortical spheroids. J Hazard Mater. 2022;435:128884. doi:10.1016/j.jhazmat.2022.128884
- Tian L, Zhang Y, Chen J, et al. Effects of nanoplastic exposure during pregnancy and lactation on neurodevelopment of rat offspring. Journal of Hazardous Materials. 2024;474:134800. doi:10.1016/j.jhazmat.2024.134800
- Zou L, Xu X, Wang Y, et al. Neonatal Exposure to Polystyrene Nanoplastics Impairs Microglia-Mediated Synaptic Pruning and Causes Social Behavioral Defects in Adulthood. Environ Sci Technol. 2024;58(27):11945-11957. doi:10.1021/acs.est.4c03231
- Chen J, Yan L, Zhang Y, et al. Maternal exposure to nanopolystyrene induces neurotoxicity in offspring through P53-mediated ferritinophagy and ferroptosis in the rat hippocampus. J Nanobiotechnology. 2024;22(1):651. doi:10.1186/s12951-024-02911-9
- de Sousa AKA, Pires KSN, Cavalcante IH, et al. Polystyrene microplastics exposition on human placental explants induces time-dependent cytotoxicity, oxidative stress and metabolic alterations. Front Endocrinol. 2024;15. doi:10.3389/fendo.2024.1481014
- Zhao HN, Thomas SP, Zylka MJ, Dorrestein PC, Hu W. Urine Excretion, Organ Distribution, and Placental Transfer of 6PPD and 6PPD-Quinone in Mice and Potential Developmental Toxicity through Nuclear Receptor Pathways. Environ Sci Technol. 2023;57(36):13429-13438. doi:10.1021/acs.est.3c05026
- Symeonides C, Vacy K, Thomson S, et al. Male autism spectrum disorder is linked to brain aromatase disruption by prenatal BPA in multimodal investigations and 10HDA ameliorates the related mouse phenotype. Nat Commun. 2024;15(1):6367. doi:10.1038/s41467-024-48897-8
- Goodrich AJ, Bennett DH, Kleeman MJ, et al. Fine and ultrafine particulate matter components and autism spectrum disorder (ASD). Environ Int. 2026;208:110068. doi:10.1016/j.envint.2026.110068
- Park E, Shen Q, Zhang Z, et al. Gestational exposure to particulate matter from urban wildfires is associated with changes in circulating oxylipins but not flame retardants 7 to 13 months post-exposure. Environ Int. 2025;200:109468. doi:10.1016/j.envint.2025.109468
- Oh J, Van de Water J, O’Connor TG, et al. Prenatal exposure to organophosphate ester flame retardants and plasticizers and maternal immune responses in three ECHO cohorts. Environ Res. 2026;305(Pt 1):124940. doi:10.1016/j.envres.2026.124940
- Selhub J, Rosenberg IH. Excessive folic acid intake and relation to adverse health outcome. Biochimie. 2016;126:71-78. doi:10.1016/j.biochi.2016.04.010
- Stefanyshyn V, Stetsyuk R, Hrebeniuk O, et al. Analysis of the Association Between the SLC19A1 Genetic Variant (rs1051266) and Autism Spectrum Disorders, Cerebral Folate Deficiency, and Clinical and Laboratory Parameters. J Mol Neurosci. 2025;75(2):42. doi:10.1007/s12031-025-02338-3
- Ramaekers VT, Sequeira JM, Quadros EV. Autoantibodies to Folate Receptors in the Cerebral Folate Deficiency Syndrome. The New England Journal of Medicine. Published online 2005.
- Gevezova M, Ivanov Z, Pacheva I, et al. Bioenergetic and Inflammatory Alterations in Regressed and Non-Regressed Patients with Autism Spectrum Disorder. International Journal of Molecular Sciences. 2024;25(15):15. doi:10.3390/ijms25158211
- Frye RE, McCarty PJ, Werner BA, Rose S, Scheck AC. Bioenergetic signatures of neurodevelopmental regression. Front Physiol. 2024;15:1306038. doi:10.3389/fphys.2024.1306038
- Che X, Roy A, Bresnahan M, et al. Metabolomic analysis of maternal mid-gestation plasma and cord blood in autism spectrum disorders. Mol Psychiatry. 2023;28(6):2355-2369. doi:10.1038/s41380-023-02051-w
- Vacy K, Thomson S, Moore A, et al. Cord blood lipid correlation network profiles are associated with subsequent attention-deficit/hyperactivity disorder and autism spectrum disorder symptoms at 2 years: a prospective birth cohort study. EBioMedicine. 2024;100:104949. doi:10.1016/j.ebiom.2023.104949
- Love C, Sominsky L, O’Hely M, Berk M, Vuillermin P, Dawson SL. Prenatal environmental risk factors for autism spectrum disorder and their potential mechanisms. BMC Med. 2024;22(1):393. doi:10.1186/s12916-024-03617-3
- Woo T, King C, Ahmed NI, et al. microRNA as a Maternal Marker for Prenatal Stress-Associated ASD, Evidence from a Murine Model. J Pers Med. 2023;13(9):1412. doi:10.3390/jpm13091412
- Subashi E, Lemaire V, Petroni V, Pietropaolo S. The Impact of Mild Chronic Stress and Maternal Experience in the Fmr1 Mouse Model of Fragile X Syndrome. Int J Mol Sci. 2023;24(14):11398. doi:10.3390/ijms241411398
- Alamoudi RA, Al-Jabri BA, Alsulami MA, Sabbagh HJ. Prenatal maternal stress and the severity of autism spectrum disorder: A cross-sectional study. Dev Psychobiol. 2023;65(2):e22369. doi:10.1002/dev.22369
- Laugesen K, Skajaa N, Petersen I, et al. Mental Disorders Among Offspring Prenatally Exposed to Systemic Glucocorticoids. JAMA Netw Open. 2025;8(1):e2453245. doi:10.1001/jamanetworkopen.2024.53245
- Cawthorpe D, Marriott B, Paget J, Moulai I, Cheung S. Relationship Between Adverse Childhood Experience Survey Items and Psychiatric Disorders. Perm J. 2018;22:18-001. doi:10.7812/TPP/18-001
- Felitti VJ, Anda RF, Nordenberg D, et al. Relationship of Childhood Abuse and Household Dysfunction to Many of the Leading Causes of Death in Adults: The Adverse Childhood Experiences (ACE) Study. American Journal of Preventive Medicine. 1998;14(4):245-258. doi:10.1016/S0749-3797(98)00017-8
- van den Oord CLJD, Copeland WE, Zhao M, Xie LY, Aberg KA, van den Oord EJCG. DNA methylation signatures of childhood trauma predict psychiatric disorders and other adverse outcomes 17 years after exposure. Mol Psychiatry. 2022;27(8):3367-3373. doi:10.1038/s41380-022-01597-5
- Ayoub G. Pollutant Burden in Autism Spectrum Disorder: Mechanistic Convergence, Genetic Susceptibility, and Clinical Translation. Current Issues in Molecular Biology. 2026;48(9):878. doi:10.3390/cimb48090878
- Zaheer J, Kim H, Ko IO, et al. Pre/post-natal exposure to microplastic as a potential risk factor for autism spectrum disorder. Environ Int. 2022;161:107121. doi:10.1016/j.envint.2022.107121
- McCanlies EC, Gu JK, Ma CC, Sanderson WT, Ludeña-Rodriguez YJ, Hertz-Picciotto I. The effects of parental occupational exposures on autism spectrum disorder severity and skills in cognitive and adaptive domains in children with autism spectrum disorder. Int J Hyg Environ Health. 2025;268:114613. doi:10.1016/j.ijheh.2025.114613
- Mamun MA, Al-Mamun F, Roy N, et al. Preconception and gestational versus postnatal exposure to air pollutants and risk of autism spectrum disorder: a systematic review and meta-analysis. Int Arch Occup Environ Health. 2025;98(1):33-57. doi:10.1007/s00420-024-02112-9
- Cloutier M, Yu C, Talarico R, et al. Prenatal Exposure to Fine Particulate Matter Components and Autism Risk in Childhood. JAMA Netw Open. 2025;8(10):e2538882. doi:10.1001/jamanetworkopen.2025.38882
- Petrick LM, Dassanayake PS, Gennings C, et al. Fetal exposure to persistent organic pollutants and childhood autism risk. Environ Res. 2026;306(Pt 2):125179. doi:10.1016/j.envres.2026.125179
- Ames JL, Ferrara A, Feng J, et al. Exposure to organophosphate ester flame retardants and plasticizers during pregnancy and autism-related outcomes in the ECHO Cohort. Environ Health Perspect. Published online June 11, 2025. doi:10.1289/EHP16177
- Chang K, Frankovich J, Cooperstock M, et al. Clinical Evaluation of Youth with Pediatric Acute-Onset Neuropsychiatric Syndrome (PANS): Recommendations from the 2013 PANS Consensus Conference. Journal of Child and Adolescent Psychopharmacology. 2015;25(1):3-13. doi:10.1089/cap.2014.0084
- Wald ER, Eickhoff J, Flood GE, et al. Estimate of the incidence of PANDAS and PANS in 3 primary care populations. Front Pediatr. 2023;11:1170379. doi:10.3389/fped.2023.1170379
- Scott O, Shi D, Andriashek D, Clark B, Goez HR. Clinical clues for autoimmunity and neuroinflammation in patients with autistic regression. Dev Med Child Neurol. 2017;59(9):947-951. doi:10.1111/dmcn.13432
- Wells L, O’Hara N, Frye RE, Hullavard N, Smith E. Folate Receptor Alpha Autoantibodies in the Pediatric Acute-Onset Neuropsychiatric Syndrome (PANS) and Pediatric Autoimmune Neuropsychiatric Disorders Associated with Streptococcal Infections (PANDAS) Population. J Pers Med. 2024;14(2):166. doi:10.3390/jpm14020166
- Antoon JW, Williams DJ, Bruce J, Sekmen M, Zhu Y, Grijalva CG. Influenza With and Without Oseltamivir Treatment and Neuropsychiatric Events Among Children and Adolescents. JAMA Neurol. 2025;82(10):1013-1021. doi:10.1001/jamaneurol.2025.1995
- Ellul P, Fourcade G, Mhanna V, et al. Maternal immune activation imprints a regulatory T cell deficiency in offspring that drives an autism-like phenotype. bioRxiv. Preprint posted online January 8, 2025:2025.01.06.631430. doi:10.1101/2025.01.06.631430
- Ellul P, Maruani A, Vantalon V, et al. Maternal immune activation during pregnancy is associated with more difficulties in socio-adaptive behaviors in autism spectrum disorder. Sci Rep. 2023;13(1):17687. doi:10.1038/s41598-023-45060-z
- Yin H, Wang Z, Liu J, et al. Dysregulation of immune and metabolism pathways in maternal immune activation induces an increased risk of autism spectrum disorders. Life Sci. 2023;324:121734. doi:10.1016/j.lfs.2023.121734
- Croen LA, Ames JL, Qian Y, et al. Inflammatory Conditions During Pregnancy and Risk of Autism and Other Neurodevelopmental Disorders. Biol Psychiatry Glob Open Sci. 2024;4(1):39-50. doi:10.1016/j.bpsgos.2023.09.008
- Bruce MR, Couch ACM, Grant S, et al. Altered behavior, brain structure, and neurometabolites in a rat model of autism-specific maternal autoantibody exposure. Mol Psychiatry. 2023;28(5):2136-2147. doi:10.1038/s41380-023-02020-3
- Ramirez-Celis A, Croen LA, Yoshida CK, et al. Maternal autoantibody profiles as biomarkers for ASD and ASD with co-occurring intellectual disability. Mol Psychiatry. 2022;27(9):3760-3767. doi:10.1038/s41380-022-01633-4
- Ramaekers VT, Sequeira JM, Blau N, Quadros EV. A milk-free diet downregulates folate receptor autoimmunity in cerebral folate deficiency syndrome. Dev Med Child Neurol. 2008;50(5):346-352. doi:10.1111/j.1469-8749.2008.02053.x
- Smith TJ, Saddiki H, Braun JM, et al. Air Pollution Mixture and Folate Status Indicators among Pregnant Women in Canada: Maternal-Infant Research on Environmental Chemicals (MIREC) Study, 2008 to 2011. Current Developments in Nutrition. 2026;10(7). doi:10.1016/j.cdnut.2025.107617
- Santoro JD, Patel L, Kammeyer R, et al. Assessment and Diagnosis of Down Syndrome Regression Disorder: International Expert Consensus. Front Neurol. 2022;13:940175. doi:10.3389/fneur.2022.940175
- Duque-Cartagena T, Dalla MDB, Mundstock E, et al. Environmental pollutants as risk factors for autism spectrum disorders: a systematic review and meta-analysis of cohort studies. BMC Public Health. 2024;24(1):2388. doi:10.1186/s12889-024-19742-w
- Gardner RM, Brynge M, Sjöqvist H, Dalman C, Karlsson H. Maternal Immune Activation and Autism in Offspring: What Is the Evidence for Causation? Biol Psychiatry. 2025;97(12):1127-1138. doi:10.1016/j.biopsych.2024.11.009
- Bai D, Yip BHK, Windham GC, et al. Association of Genetic and Environmental Factors With Autism in a 5-Country Cohort. JAMA Psychiatry. 2019;76(10):1035-1043. doi:10.1001/jamapsychiatry.2019.1411
- Huang RFS, Hsu YC, Lin HL, Yang FL. Folate Depletion and Elevated Plasma Homocysteine Promote Oxidative Stress in Rat Livers. The Journal of Nutrition. 2001;131(1):33-38. doi:10.1093/jn/131.1.33
- Antoniades C, Shirodaria C, Warrick N, et al. Global improvement of vascular function and redox state with low-dose folic acid: implications for folate therapy in patients with coronary artery disease. Circulation. 2007;115:2262-2270.
- Engevik MA, Morra CN, Röth D, et al. Microbial Metabolic Capacity for Intestinal Folate Production and Modulation of Host Folate Receptors. Front Microbiol. 2019;10:2305. doi:10.3389/fmicb.2019.02305
- Michel A, Pourié G, Kökten T. Consequences of Western and Mediterranean Diets’ Nutrients on the Microbiota-Gut-Brain Axis. Nutrients. 2026;18(8):1258. doi:10.3390/nu18081258
- Plaza-Díaz J, Gómez-Fernández A, Chueca N, et al. Autism Spectrum Disorder (ASD) with and without Mental Regression is Associated with Changes in the Fecal Microbiota. Nutrients. 2019;11(2):337. doi:10.3390/nu11020337
- Sotelo-Orozco J, Taft DH, Al-Oboudi J, et al. Mixed Evidence for Impact of Early Infant Gut Microbiome and Later Development of Autism Spectrum Disorder in the MARBLES Prospective Cohort Study. Autism Research. 2026;19(4):e70207. doi:10.1002/aur.70207
- Yap CX, Henders AK, Alvares GA, et al. Autism-related dietary preferences mediate autism-gut microbiome associations. Cell. 2021;184(24):5916-5931.e17. doi:10.1016/j.cell.2021.10.015
- Kerns CM, Zaidman-Zait A, Robins DL, et al. Psychometric properties of the Childhood Adversity and Social Stress Questionnaire, Parent Report (CASSQ-PR), a measure tailored to assess stressful experiences and traumatic stress symptoms in autistic youth. Psychol Trauma. Published online September 15, 2025. doi:10.1037/tra0002007
- Choe JY, Nair M, Basha R, Kim BJ, Jones HP. Defining Early Life Stress as a Precursor for Autoimmune Disease. Crit Rev Immunol. 2019;39(5):329-342. doi:10.1615/CritRevImmunol.2020033244
- Desplats P, Gutierrez AM, Antonelli MC, Frasch MG. Microglial memory of early life stress and inflammation: susceptibility to neurodegeneration in adulthood. Neurosci Biobehav Rev. 2020;117:232-242. doi:10.1016/j.neubiorev.2019.10.013
- Catale C, Carola V, Viscomi MT. Early life stress-induced neuroinflammation and neurological disorders: a novel perspective for research. Neural Regen Res. 2022;17(9):1971-1972. doi:10.4103/1673-5374.335152
- Faraji J, Bettenson D, Yong VW, Metz GAS. Early life stress aggravates disease pathogenesis in mice with experimental autoimmune encephalomyelitis: Support for a two-hit hypothesis of multiple sclerosis etiology. Journal of Neuroimmunology. 2023;385:578240. doi:10.1016/j.jneuroim.2023.578240
- Ahmavaara K, Ayoub G. Stress and Folate Impact Neurodevelopmental Disorders. J Health Care and Research. 2024;5(1):1-6. doi:10.36502/2024/hcr.6228
- Diagnosis of Autism Occurring Earlier in Children, Though Still Late for Many; Initial Diagnosis in Adulthood Increasing in Women. August 12, 2025. Accessed August 12, 2026. https://www.epicresearch.org/articles/diagnosis-of-autism-occurring-earlier-in-children-though-still-late-for-many-initial-diagnosis-in-adulthood-increasing-in-women/
- CDC. Autism and Developmental Disabilities Monitoring (ADDM) Network. Autism Spectrum Disorder (ASD). June 10, 2026. Accessed August 12, 2026. https://www.cdc.gov/autism/addm-network/index.html
- Fyfe C, Winell H, Dougherty J, et al. Time trends in the male to female ratio for autism incidence: population based, prospectively collected, birth cohort study. BMJ. 2026;392:e084164. doi:10.1136/bmj-2025-084164
- Loomes R, Hull L, Mandy WPL. What Is the Male-to-Female Ratio in Autism Spectrum Disorder? A Systematic Review and Meta-Analysis. J Am Acad Child Adolesc Psychiatry. 2017;56(6):466-474. doi:10.1016/j.jaac.2017.03.013
- Hull L, Lai MC, Baron-Cohen S, et al. Gender differences in self-reported camouflaging in autistic and non-autistic adults. Autism. 2020;24(2):352-363. doi:10.1177/1362361319864804
- Robinson EB, Lichtenstein P, Anckarsäter H, Happé F, Ronald A. Examining and interpreting the female protective effect against autistic behavior. Proc Natl Acad Sci U S A. 2013;110(13):5258-5262. doi:10.1073/pnas.1211070110
- Jacquemont S, Coe BP, Hersch M, et al. A higher mutational burden in females supports a “female protective model” in neurodevelopmental disorders. Am J Hum Genet. 2014;94(3):415-425. doi:10.1016/j.ajhg.2014.02.001
- Brister D, Rose S, Delhey L, et al. Metabolomic Signatures of Autism Spectrum Disorder. J Pers Med. 2022;12(10):1727. doi:10.3390/jpm12101727
- Edalati S, Bagherzadeh F, Jafarabadi MA, Ebrahimi-Mamaghani M. Higher ultra-processed food intake is associated with higher DNA damage in healthy adolescents. British Journal of Nutrition. 2021;125(5):568-576. doi:10.1017/S0007114520001981
- Lane MM, Lotfaliany M, Forbes M, et al. Higher Ultra-Processed Food Consumption Is Associated with Greater High-Sensitivity C-Reactive Protein Concentration in Adults: Cross-Sectional Results from the Melbourne Collaborative Cohort Study. Nutrients. 2022;14(16):3309. doi:10.3390/nu14163309
Article information
Authors
- George AyoubSanta Barbara City CollegeCorresponding author · [email protected]
How to cite
Ayoub, G., 2026. Autism Emergence Timeline: Oxidative Burden and Buffering Capacity Influence the Timing of Late Diagnosis and Regression. Medical Research Archives, [online] 14(10). https://doi.org/10.18103/mra.2026.0701
Record
- Journal
- Medical Research Archives
- Issue
- Vol 14 No 10 (2026)
- Section
- Review Articles
- Published
- 31 October 2026
- ISSN
- 2375-1924
- Downloads
- 74
Copyright and licence
Published by the European Society of Medicine under a Creative Commons Attribution 4.0 International licence (CC BY 4.0), which permits unrestricted use, distribution and reproduction in any medium, provided the original authors and source are credited.