Muscle-brain endocrine axis in autism spectrum disorder: Mechanisms, evidence, and research priorities

Abstract

Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental condition associated with alterations in synaptic plasticity, immune regulation, energy metabolism, and stress responsivity. Skeletal muscle is an endocrine organ that releases exercise-responsive mediators capable of participating in muscle-brain communication. This narrative review evaluates how irisin-related signaling, interleukin-6, brain-derived neurotrophic factor (BDNF)-related pathways, and metabolic mediators may intersect with four ASD-relevant domains: Synaptic plasticity, neuroinflammation, mitochondrial and energy metabolism, and hypothalamic-pituitary-adrenal axis regulation. ASD-specific mechanistic evidence remains limited. Gain- and loss-of-function studies in non-ASD models support the biological plausibility of several pathways, particularly FNDC5/irisin signaling, but do not establish causality in ASD. Human studies provide preliminary and heterogeneous evidence that exercise may alter peripheral biomarkers and selected behavioral or cognitive outcomes. The apparent BDNF paradox, elevated peripheral BDNF in some ASD cohorts despite the proposed benefits of exercise-related BDNF signaling, calls for compartment-, baseline-, and time-dependent interpretation. We therefore present the muscle-brain axis as a hypothesis-generating framework for future stratified trials rather than as a validated clinical prescription. Myokine profiling and pathway-targeted therapies should remain investigational pending the development of standardized assays, ASD-specific causal studies, and adequately powered randomized trials.

Key Words: Autism spectrum disorder; Skeletal muscle; Muscle-brain axis; Myokines; Irisin; Brain-derived neurotrophic factor; Neuroplasticity; Exercise intervention

Core Tip: This review evaluates skeletal muscle-brain communication as a hypothesis-generating framework for autism spectrum disorder. It distinguishes autism-specific evidence from mechanistic findings derived from other disease models, addresses the brain-derived neurotrophic factor paradox, and emphasizes that the proposed links among irisin, interleukin-6, neuroplasticity, neuroinflammation, metabolism, and stress regulation vary substantially in evidential strength. Current data do not justify biomarker-guided exercise prescriptions or myokine-targeted therapy in clinical practice. Instead, the framework identifies testable mechanisms, major translational gaps, and priorities for stratified randomized trials.

INTRODUCTION

Autism spectrum disorder (ASD) is a multifaceted neurodevelopmental condition characterized by substantial challenges in social communication and restricted, repetitive behaviors and interests. The population-level burden is considerable. The Global Burden of Disease Study 2021 estimated that approximately 61.8 million people worldwide, about one in 127, were autistic, with an estimated 11.5 million disability-adjusted life-years attributable to autism globally[1]. In the United States, 2022 surveillance across 16 Autism and Developmental Disabilities Monitoring Network sites estimated ASD prevalence at 32.2 per 1000 children aged 8 years, equivalent to approximately one in 31[2]. Traditional intervention strategies for ASD have focused on behavioral modification and skill training. However, increasing evidence indicates that the pathophysiological basis of ASD extends beyond the central nervous system (CNS) and may involve systemic dysregulation. Reported abnormalities include altered synaptic plasticity, persistent neuroinflammation, disrupted brain energy metabolism, and dysfunction of the hypothalamic-pituitary-adrenal (HPA) axis stress response[3-5]. From this systemic perspective, it is reasonable to investigate interventions that may influence interacting physiological systems while recognizing that the causal relevance of these abnormalities to ASD symptoms varies among individuals and remains incompletely resolved.

Regular physical activity has been associated with benefits in social skills, executive function, and emotional and behavioral problems in individuals with ASD[6]. However, its mechanisms of action have long been attributed primarily to psychosocial and behavioral effects. Recent evidence has broadened this perspective. In particular, skeletal muscle is not merely an effector organ for movement but also an active endocrine organ[7]. During contraction, skeletal muscle releases or modulates a range of exercise-responsive mediators. These include established or candidate myokines, such as interleukin-6 (IL-6), and irisin-related signaling, whereas brain-derived neurotrophic factor (BDNF) requires more careful interpretation because muscle-derived BDNF appears to act predominantly in an autocrine/paracrine manner, and circulating BDNF has multiple tissue sources[8,9]. Throughout this review, the broader term “exercise-responsive mediators” is used to refer to a mixed class that includes myokines, neurotrophic factors, and metabolites such as lactate. Together, these signals may participate in muscle-brain communication through circulatory, neural, vascular, and metabolic routes[10].

This review assesses how the emerging muscle-brain axis framework provides conceptual and translational perspectives for understanding and modulating neuroplasticity in ASD. First, we outline the biosynthesis and release of major myokines and the principal routes through which they may influence the brain and exert biological effects. We then assess the intersections between myokine signaling and four major pathophysiological domains of ASD: Dysfunction of synaptic plasticity and neurotrophic support; neuroinflammation and microglial dysfunction; energy metabolism and mitochondrial dysregulation; and HPA axis and stress-response dysfunction. These intersections are supported by evidence of varying maturity. Experimental studies suggest that FNDC5/irisin-related signaling can influence hippocampal BDNF-associated pathways[11], whereas the biological effects of exercise-related IL-6 are strongly context dependent and vary according to signaling mode and exercise conditions[12]. Direct evidence that exercise-induced peripheral IL-6 reprograms human microglia in ASD is currently lacking. Likewise, lactate is an exercise-responsive metabolite rather than a myokine. However, it may serve as an oxidative substrate and signaling molecule in the brain[13]; its capacity to compensate for ASD-associated mitochondrial dysfunction remains unproven[5,13]. Accordingly, the following sections distinguish established associations, indirect mechanistic support, and hypothesis-level extrapolations.

On this basis, we synthesize ASD-specific preclinical and human evidence separately from transdiagnostic mechanistic studies conducted in non-ASD models. We then consider whether genotype, endophenotype, and behavioral preferences could eventually inform stratified exercise research. Candidate myokines and related mediators are discussed as exploratory response measures rather than validated biomarkers, whereas recombinant irisin and tropomyosin receptor kinase B (TrkB)-directed agents are considered investigational hypotheses rather than established therapeutic options[14-16]. Finally, we examine potential combinations of exercise with behavioral therapy, neuromodulation, and pharmacotherapy while explicitly distinguishing proposed synergy from demonstrated ASD-specific efficacy.

We therefore propose the muscle-brain axis as an integrative framework for organizing several biological hypotheses concerning exercise and ASD. The framework extends the view of exercise beyond mere behavioral participation. However, it should not be interpreted as evidence that skeletal muscle endocrine signaling is an established treatment mechanism for ASD. Its principal value lies in generating testable questions about neuroplasticity, immune regulation, metabolism, and neuroendocrine function and identifying areas in which ASD-specific causal evidence is still required.

METHODOLOGY

This structured narrative review synthesized mechanistic, preclinical, and clinical literature on exercise-responsive muscle signaling and its potential relevance to ASD. The aim was to develop a conceptually coherent, hypothesis-generating framework rather than to conduct a systematic review, meta-analysis, or clinical guideline assessment. Literature published through March 2026 was searched in PubMed/MEDLINE, Web of Science Core Collection, and Scopus using combinations of terms related to ASD, skeletal muscle, myokines, exercise, irisin, BDNF, IL-6, neuroplasticity, neuroinflammation, mitochondrial dysfunction, metabolism, and stress responses. The reference lists of relevant articles and reviews were also screened. Eligible literature included original mechanistic studies, animal experiments, human observational or interventional studies, and reviews that contributed directly to at least one of four themes: Skeletal muscle endocrine function; exercise-responsive mediators and muscle-brain communication; ASD-relevant synaptic, immune, metabolic, or stress-response biology; or the translational implications of exercise-based interventions.

To avoid conflating levels of evidence, studies were interpreted in three categories: (1) ASD-specific evidence from ASD models or autistic participants; (2) Transdiagnostic mechanistic evidence from non-ASD experimental models that supports biological plausibility but not ASD-specific causality; and (3) Translational hypotheses requiring prospective validation. Because study designs, populations, exercise protocols, biomarker assays, and outcomes were highly heterogeneous, no quantitative synthesis or formal risk-of-bias assessment was performed. Accordingly, proposed biomarkers, precision-exercise strategies, recombinant myokines, and pathway-targeted agents are discussed as research directions rather than as validated clinical recommendations.

FINDINGS

Myokines: Molecular messengers from skeletal muscle to the brain

Definition and discovery: Myokines are cytokines or peptide factors produced by skeletal muscle cells during contraction, such as that occurring during exercise, or in response to other biological signals[7]. As important molecular messengers, they not only modulate skeletal muscle metabolism, growth, and repair through autocrine and paracrine mechanisms but also communicate with distant organs, including adipose tissue, the liver, bone, and the brain, through endocrine signaling via the systemic circulation. This activity constitutes a complex network of muscle-organ crosstalk through which myokines play important roles in systemic energy homeostasis, immune modulation, and neurological function[8,17]. Their discovery has fundamentally challenged the traditional view of skeletal muscle as solely an organ of locomotion and has established its important endocrine role.

Advances in secretomic research have identified more than 600 putative myokines[8]. Among them, several major myokines have attracted substantial attention because of their effects on the nervous system, supporting a molecular basis for the muscle-brain axis. Irisin is generated through the exercise-induced skeletal muscle peroxisome proliferator-activated receptor γ coactivator-1α/FNDC5 pathway[18]. It promotes the browning of white adipose tissue and influences systemic energy metabolism[19]. Peripheral FNDC5/irisin signaling has also been associated with brain effects in experimental models, including changes in hippocampal BDNF-related pathways[20]. However, the extent, route, and physiological significance of the direct transport of circulating irisin across the human blood-brain barrier remain incompletely resolved; brain effects may also arise through indirect vascular, neural, or metabolic mechanisms. Dysregulation of this signaling pathway has been associated with various neurodegenerative disorders[21]. Notably, BDNF is also produced by skeletal muscle during contraction. Although muscle-derived BDNF may not directly reach the brain, it can influence the CNS indirectly through peripheral actions. Exercise-induced increases in cerebral BDNF are often associated with myokine signaling, including irisin-related signaling[22,23]. IL-6 is among the first myokines to increase substantially and be released during physical activity, primarily acting through the classical signaling pathway. Conversely, under chronic pathological conditions, IL-6 produced by immune cells via the trans-signaling pathway primarily promotes inflammatory responses[24,25]. Exercise also influences circulating levels of insulin-like growth factor 1 (IGF-1). By activating its receptor and downstream pathways such as phosphoinositide 3-kinases/protein kinase B, IGF-1 plays a crucial role in the proliferation, differentiation, and hypertrophy of muscle satellite cells[26] and also supports neuronal survival, synaptogenesis, and neurogenesis. Evidence suggests that a single session of endurance or resistance exercise can alter total serum IGF-1 levels[27]. After exercise, the mRNA expression of leukemia inhibitory factor, a member of the IL-6 cytokine family, increases in human skeletal muscle[28]. Research indicates that leukemia inhibitory factor may exert neurotrophic effects on neural progenitor cells derived from human pluripotent stem cells by improving survival, promoting differentiation, and reducing apoptosis[29].

Overall, skeletal muscle contributes to a broader network of exercise-responsive signaling that may influence the CNS directly or indirectly. Myokines are candidate components of this network, but they should not be assumed to be the primary mediators of all exercise-related brain effects, as circulating signals arise from multiple tissues and include non-myokine metabolites and neuroendocrine factors. This distinction is particularly important in ASD, for which direct causal evidence linking specific muscle-derived factors to behavioral outcomes remains limited.

Biosynthesis and release pathways of major myokines: Exercise mode, intensity, and duration substantially affect the synthesis and release of major myokines. The biosynthetic pathway of irisin is well characterized. Exercise first activates peroxisome proliferator-activated receptor γ coactivator-1α in skeletal muscle. This master regulator of metabolism upregulates the expression of the transmembrane protein FNDC5[11]. The extracellular domain of FNDC5 is enzymatically cleaved, releasing the active form, irisin, into the bloodstream[30]. Different exercise modalities, including resistance training, aerobic exercise, and high-intensity interval training, can increase irisin levels[31]. By activating multiple signaling pathways, including mitogen-activated protein kinases and AMP-activated protein kinase, irisin not only promotes the browning of white adipose tissue to increase thermogenesis[19] but also functions as an important endocrine signal. Experimental evidence links peripheral FNDC5/irisin signaling to BDNF-related pathways in the brain[11], making it a candidate molecular bridge between exercise and neuroplasticity; however, the relative contributions of direct blood-brain barrier transport and indirect signaling remain unresolved.

BDNF itself is also synthesized by skeletal muscle cells during exercise, but its mode of action differs from that of centrally derived BDNF. Muscle contraction, whether simulated by electrical stimulation or induced by exercise, can broadly increase BDNF mRNA and protein expression in skeletal muscle. However, muscle-derived BDNF primarily functions as an autocrine/paracrine factor within muscle tissue, enhancing lipid oxidation by activating AMP-activated protein kinase signaling, and is not released into the bloodstream in substantial amounts[9]. It plays an important role in post-exercise recovery by facilitating functional adaptation and supporting muscle endurance through peroxisome proliferator-activated receptor delta-dependent metabolic reprogramming[32]. In addition, during development, muscle-derived BDNF plays a crucial role in the early formation of postsynaptic structures at the neuromuscular junction[33].

Other important myokines exhibit distinct release dynamics and functions. IL-6 is particularly responsive to exercise, though its effects are highly context-dependent. A single bout of high-intensity exercise can markedly increase plasma IL-6 levels and may be accompanied by pro-inflammatory responses, such as increased neutrophil migration[34,35]. Conversely, long-term, regular, moderate-intensity exercise training promotes a systemic anti-inflammatory state. In this context, IL-6 may exert metabolic regulatory and context-dependent neuroprotective effects through its classical signaling pathway[36,37]. Myostatin, a member of the transforming growth factor-β superfamily, is a major negative regulator of muscle growth[38]. Its expression is upregulated during muscle injury and in certain pathological states, thereby inhibiting muscle satellite cell activation and muscle protein synthesis[39,40]. Notably, the endoplasmic reticulum stress sensor IRE1α can promote myostatin mRNA decay, thereby positively regulating muscle regeneration[41]. This finding illustrates the precise balancing mechanisms within the myokine network.

Overall, skeletal muscle dynamically secretes biologically active myokines through distinct molecular pathways. These factors either modulate distant organs, such as the brain, or precisely regulate muscle metabolism, repair, and growth through autocrine/paracrine mechanisms. Collectively, they contribute to the molecular basis of the systemic benefits of exercise.

Mechanisms of myokine entry and action in the brain: Peripheral signaling molecules synthesized by skeletal muscle influence brain function through a multilayered communication network rather than through a single pathway. This network primarily involves three mechanisms: Direct transport across the blood-brain barrier, signal relay through the vagus nerve (VN), and indirect effects on the CNS through modulation of systemic metabolic status[10].

Direct access to the CNS is one possible route for peripheral exercise-responsive signals, but this mechanism should not be generalized across mediators. Experimental studies have linked peripheral FNDC5/irisin signaling to central effects, including BDNF-related changes[42]. In mice, peripherally administered irisin has been shown to cross the blood-brain barrier and to influence cognitive function[42]. However, evidence for a corresponding transport process in humans remains limited. Irisin has been detected and quantified in human cerebrospinal fluid by tandem mass spectrometry. However, the route of transport from blood to cerebrospinal fluid and the quantitative contribution of circulating irisin to human brain exposure remain unresolved[43]. At present, no specific receptor-mediated transport mechanism for irisin across the human blood-brain barrier has been established. Accordingly, direct transport, endothelial signaling, neural relay, and systemic metabolic effects should be treated as distinct and incompletely resolved possibilities. In addition, human studies show that exercise can alter circulating concentrations of BDNF, IL-6, irisin, and other exercise-responsive factors in some settings, but responses vary according to exercise mode, intensity, and duration; participant characteristics; sampling conditions; and assay methodology[12,44,45]. Moreover, circulating BDNF has multiple potential sources and is influenced by release and storage mechanisms outside skeletal muscle[44]. Consequently, a change in peripheral concentration cannot, by itself, establish a skeletal muscle origin, central target engagement, or mediation of behavioral change. These limitations are central to interpreting the proposed muscle-brain axis in ASD.

Signal relay through the VN constitutes another important neurohumoral pathway. The VN is the primary parasympathetic nerve connecting peripheral organs to the CNS and conveys information about their physiological state to regions such as the nucleus tractus solitarius in the brainstem. Metabolic changes resulting from skeletal muscle contraction or the production of specific factors such as IL-6 during exercise may activate vagal afferent fibers. These fibers subsequently transmit signals to the brain, altering the activity of neurotransmitter systems, including the serotonergic and dopaminergic systems, and thereby affecting mood, cognition, and behavior[46]. Although myokines have been proposed to influence the brain through this pathway, the specific myokines that serve as primary vagal activators and the precise mechanisms of signal transmission remain to be elucidated[47].

A further mechanism involves the indirect modulation of brain function through changes in peripheral metabolism. By releasing myokines such as irisin, IL-6, and FGF21, skeletal muscle contributes to the regulation of whole-body energy balance, enhancement of insulin sensitivity, reduction of hepatic glucose production, and modulation of adipose tissue activity[48]. As an organ with high energy demands, the brain depends on a consistent and efficient supply of energy substrates, primarily glucose. Myokines may improve systemic metabolism, thereby helping ensure an adequate supply of energy and essential metabolic intermediates to the brain. These substrates support core processes such as synaptic transmission and neurotransmitter synthesis, which provide a foundation for neuroplasticity[49]. This indirect mechanism may help maintain brain health in the presence of metabolic dysregulation, such as insulin resistance. However, variations in circulating levels of certain exerkines, including BDNF and irisin, may reflect contributions from multiple organs, such as the brain and adipose tissue[10]. Therefore, causal relationships between these circulating factors and brain function should be inferred cautiously.

Experimental studies provide the strongest mechanistic support for exercise-responsive muscle-brain signaling. Genetic, pharmacological, and molecular manipulations have linked FNDC5/irisin-related pathways to selected neural, behavioral, and peripheral metabolic outcomes[42,50-52]. However, much of this evidence is derived from healthy animals, aging models, neurodegenerative disease models, or other non-ASD conditions[42,50-52]. These studies support biological plausibility but should not be interpreted as evidence that the same pathways causally mediate exercise responses in ASD.

Intersections of major ASD pathophysiology and myokine signaling

Myokine signaling does not operate through a single mechanism but may intersect with multiple key pathophysiological domains of ASD. The following sections examine how exercise-responsive peripheral mediators may intersect with synaptic plasticity, neuroinflammation, mitochondrial and energy metabolism, and HPA axis function[3-5,53,54]. The evidential strength varies substantially across these proposed pathways, and direct ASD-specific mediation should not be inferred from mechanistic findings in other conditions. Table 1 summarizes the current evidence across these mechanistic domains, distinguishing ASD-specific findings from transdiagnostic mechanistic evidence and human clinical observations while highlighting the principal methodological and translational gaps. Figure 1 should be interpreted as an evidence-graded conceptual map rather than as a set of established causal pathways. The proposed BDNF/TrkB link is supported mainly by experimental studies and limited human biomarker data; the neuroimmune pathway is more indirect because neither the direct transport of muscle-derived IL-6 into the human ASD brain nor the causal reprogramming of microglia has been demonstrated; lactate-related mechanisms involve a metabolic mediator rather than a myokine and remain context dependent; and HPA-axis effects are inferred mainly from exercise–stress associations rather than demonstrated myokine mediation in ASD. These differences in evidential strength are represented graphically by distinct arrow styles.

Figure 1 Intersections between major autism spectrum disorder pathophysiological domains and myokine signaling. Proposed intersections between exercise-responsive peripheral mediators and autism spectrum disorder-relevant pathophysiological domains. Solid arrows denote relatively well-supported associations, dashed arrows denote indirect links, and dotted arrows denote hypothesis-level relationships requiring autism-specific causal validation. Candidate pathways include brain-derived neurotrophic factor/tropomyosin receptor kinase B-related plasticity, neuroimmune signaling, lactate transport and metabolism, and hypothalamic-pituitary-adrenal-axis regulation. The figure should not be interpreted as evidence that muscle-derived interleukin-6 directly crosses the blood-brain barrier or causes a binary M1-to-M2 microglial switch in autism spectrum disorder. MR: Mineralocorticoid receptor; AMPK: AMP-activated protein kinase; PGC-1α: Peroxisome proliferator-activated receptor γ coactivator-1α; FNDC5: Fibronectin type III domain-containing protein 5; IL: Interleukin; MCT: Monocarboxylate transporter; ASD: Autism spectrum disorder; TrkB: Tropomyosin receptor kinase B; HPA: Hypothalamic-pituitary-adrenal; ANLS: Astrocyte-neuron lactate shuttle.

Table 1 Evidence map and translational gaps across proposed muscle-brain mechanisms in autism spectrum disorder.

Mechanistic domain	ASD-specific evidence	Transdiagnostic mechanistic evidence	Human/clinical evidence	Typical methods represented	Current interpretation and key translational gap	Ref.
Synaptic plasticity and BDNF/TrkB signaling	Altered synaptic structure and function are reported in ASD-relevant models and subgroups. Peripheral BDNF findings are heterogeneous; MDGA2 deficiency can produce excessive BDNF/TrkB activation and autism-relevant phenotypes in mice. An ASD exercise trial reported concurrent serum BDNF and executive-function changes	Exercise-related PGC-1α/FNDC5/irisin signaling can influence hippocampal BDNF-associated pathways in non-ASD systems. Mechanistic studies also show that precursor and mature BDNF can engage distinct signaling programs	Peripheral BDNF differs between some ASD and comparison cohorts, but serum/plasma measures are strongly influenced by platelet storage and release and by pre-analytical handling. Parallel changes in biomarkers and behavior do not establish mediation	Genetic and pathway-manipulation models; systematic review/meta-analysis; peripheral biomarker assays; exercise intervention trial	Biological plausibility is substantial, but ASD-specific causal mediation remains unproven. The BDNF paradox requires baseline-, compartment-, time-, and molecular-form-specific testing; peripheral BDNF should remain an exploratory marker of response	[3,11,55-65]
Neuroinflammation and microglial states	Innate immune dysfunction, neuroinflammation, and altered microglial programs have been implicated in subsets of ASD. Causal role, developmental timing, and subgroup specificity remain unresolved	Exercise-related IL-6 responses are context dependent. Human experimental studies show that IL-6 can increase IL-1ra/IL-10 and that exercise or IL-6 infusion can attenuate endotoxin-induced TNF-α. Single-cell studies demonstrate multidimensional microglial heterogeneity beyond a binary M1/M2 model	There is no direct demonstration that exercise-induced, muscle-derived IL-6 crosses the blood-brain barrier and reprograms human microglia in ASD. Clinical ASD evidence is therefore indirect	ASD neuroimmune reviews; human cytokine infusion/exercise experiments; single-cell transcriptomics; observational immune profiling	The pathway is best treated as an indirect, hypothesis-level model. A demonstrated M1-to-M2 conversion should not be inferred. ASD studies need time-resolved peripheral mediator measures linked to central neuroimmune readouts and functional outcomes	[4,12,66-70]
Energy metabolism, mitochondrial dysfunction, and lactate	Mitochondrial abnormalities and mitochondrial-dysfunction biomarkers have been reported in ASD subsets, with substantial biological heterogeneity. A single cerebral-energy-deficit model should not be generalized across the spectrum	Exercise improves insulin signaling and systemic glucose regulation. Lactate acts as a metabolic substrate and signaling molecule; the human brain can increase lactate uptake during exercise. Sustained oxidative use of lactate depends on downstream mitochondrial capacity	Direct human evidence that exercise-derived lactate corrects ASD-associated mitochondrial dysfunction or mediates behavioral improvement is insufficient	Systematic reviews/meta-analyses; metabolic physiology; brain substrate-uptake studies; mechanistic and integrative reviews	Metabolic support is plausible, but lactate is not a myokine and should not be presented as a generic bypass for mitochondrial dysfunction. ASD-specific mediation requires metabolic phenotyping linked to brain measures and outcomes	[5,13,71-79]
HPA axis and stress response	Meta-analytic and review evidence indicates altered cortisol patterns and stress responsivity in ASD, but findings vary by developmental stage, sampling method, stress paradigm, and subgroup. Exercise studies report selected improvements in emotional and behavioral outcomes	In broader populations, higher physical activity is associated with a steeper diurnal cortisol slope, although effects are not uniform across HPA indices. IL-6- and irisin-related immune/neurotrophic pathways provide only indirect links to stress-regulatory systems	No direct evidence establishes that a specific muscle-derived myokine recalibrates HPA-axis function in ASD. Behavioral change can arise through multiple nonexclusive mechanisms	Cortisol meta-analysis; stress-system review; physical-activity meta-analysis; ASD behavioral intervention studies	The muscle-HPA link is a testable indirect hypothesis, not an established mechanism. Future trials require repeated cortisol sampling, standardized stress tasks, mediator kinetics, and formal mediation analyses	[6,53,54,60,68,80-82]
Cross-cutting biomarkers and stratified exercise translation	ASD exercise outcomes are heterogeneous. Broad symptom severity and stereotyped behaviors show null or inconsistent effects in some syntheses. Peripheral BDNF and irisin profiles vary across cohorts, and an aquatic-exercise trial showed concurrent, but not necessarily mediated, changes in BDNF and executive function	Exercise-responsive BDNF, IL-6, and irisin vary with intervention characteristics, participant characteristics, sampling conditions, and assay methodology. Multiple non-muscle factors influence circulating BDNF levels; human irisin studies show methodological heterogeneity	No validated ASD-specific biomarker thresholds, response-prediction rules, or genotype- or omics-based exercise decision algorithms currently exist. Pediatric BDNF RCT evidence is not uniformly positive and is not ASD-specific	Clinical-trial syntheses; systematic reviews/meta-analyses; biomarker studies; exercise RCTs; assay-methodology reviews	Myokines and related mediators should be treated as candidate research measures, not validated clinical biomarkers. Priorities include assay standardization, prespecified subgroup analyses, mediation testing, external validation, and comparison with standard clinical assessment	[6,12,44,45,57,59,62-64,86-90,97,98,102]

Evidence categories reflect study context and narrative interpretation, not formal risk-of-bias grading.

ASD: Autism spectrum disorder; BDNF: Brain-derived neurotrophic factor; TrkB: Tropomyosin receptor kinase B; PGC-1α: Peroxisome proliferator-activated receptor γ coactivator-1α; FNDC5: Fibronectin type III domain-containing protein 5; IL: Interleukin; TNF-α: Tumor necrosis factor-α; HPA: Hypothalamic-pituitary-adrenal; RCT: Randomized controlled trial.

Dysregulation of synaptic plasticity and neurotrophic support: Synaptic plasticity is central to experience-dependent learning and circuit refinement, and altered synaptic structure and function have been reported across multiple ASD-relevant models and biological subgroups[3]. The BDNF/TrkB pathway is an important regulator of neuronal survival, dendritic remodeling, synaptic maturation, and long-term potentiation through downstream pathways that include mitogen-activated protein kinase, phosphoinositide 3-kinase, and phospholipase Cγ signaling[55,56]. However, ASD does not appear to involve a uniform deficiency in BDNF signaling. Rather, the available literature suggests heterogeneous, potentially bidirectional abnormalities across biological subgroups[57-59].

Preclinical evidence: Experimental studies indicate that deviations in neurotrophic signaling may be maladaptive and that ASD-related abnormalities cannot be reduced to a simple deficiency model[55,56,58]. For example, disruption of ASD-related synaptic regulators can alter TrkB pathway activity. Loss of MDGA2-mediated restraint has been linked to excessive BDNF/TrkB signaling, altered excitatory synaptic function, and autism-relevant behavioral phenotypes in mice[58]. This finding is important because it demonstrates that enhanced neurotrophic signaling is not necessarily beneficial and that excessive pathway activation may itself contribute to circuit imbalance. Separately, exercise and FNDC5/irisin-related signaling have been shown to influence hippocampal BDNF-associated pathways in non-ASD experimental systems[11,60,61]. These studies provide biological plausibility for an exercise-responsive peripheral-to-central pathway but do not establish that increasing BDNF is uniformly beneficial in ASD or demonstrate that muscle-derived signals normalize BDNF/TrkB dysfunction in ASD-specific models. Accordingly, evidence from non-ASD experimental systems should be interpreted as mechanistic support for hypothesis generation rather than as direct proof of causal mediation in ASD.

Clinical evidence and translational gap: Human evidence is heterogeneous. Peripheral BDNF concentrations differ between some ASD and comparison cohorts[57,59], and one exercise intervention reported concurrent increases in serum BDNF and improvements in selected executive outcomes in children with ASD[62]. However, peripheral BDNF measurements are strongly influenced by platelet storage and release dynamics and by pre-analytical conditions, including sample type, storage, and handling procedures[63,64]. Consequently, peripheral BDNF concentrations cannot be assumed to provide a direct readout of skeletal muscle secretion or central BDNF activity. The co-occurrence of increased serum BDNF and improved cognitive outcomes after exercise[62] also does not establish mediation. Such parallel changes may reflect independent responses to exercise, a shared upstream process, or a compensatory response rather than a causal pathway linking peripheral BDNF to behavioral improvement. Exercise-associated changes in circulating BDNF should therefore be interpreted as exploratory correlates rather than as evidence that a muscle-derived BDNF pathway has corrected central synaptic dysfunction.

The BDNF paradox in ASD: The BDNF literature in ASD does not support a simple deficiency model. Some cohorts show elevated peripheral BDNF concentrations associated with greater symptom severity[59], whereas mechanistic studies indicate that excessive TrkB activation can itself disrupt synaptic balance[58]. This finding creates an important paradox for exercise-based hypotheses: If BDNF-related signaling is already elevated or dysregulated in some ASD subgroups, further enhancement of this pathway may not be beneficial. Peripheral serum BDNF cannot be assumed to reflect central BDNF activity because circulating measurements are substantially influenced by platelet storage and release[63] and by pre-analytical conditions[64]. Moreover, biologically distinct molecular forms should not be conflated. Mature BDNF primarily engages TrkB-associated trophic and potentiation-related signaling[55,56], whereas precursor BDNF can signal through p75 neurotrophin receptor-related pathways and facilitate hippocampal long-term depression[65]. Therefore, an exercise-associated increase in total serum BDNF[62] should not automatically be interpreted as normalization of deficient central neurotrophic signaling.

Several testable hypotheses may explain this apparent paradox. First, exercise responses may be baseline dependent, with benefits occurring mainly in individuals with relatively low or dysregulated BDNF signaling and little benefit, or potential harm, in subgroups with already excessive pathway activity. Second, transient exercise-induced changes may differ biologically from chronically elevated resting concentrations. Third, peripheral and central BDNF trajectories may not be equivalent, particularly because circulating measurements are strongly influenced by platelet-related and pre-analytical factors. Fourth, molecular form-specific effects, including distinctions between precursor and mature BDNF signaling, warrant direct investigation[65].

Future trials should therefore stratify participants by baseline BDNF status where feasible, standardize serum or plasma sampling and processing, examine longitudinal trajectories and molecular forms, and formally test whether biomarker changes mediate rather than merely accompany functional outcomes. Until such evidence is available, peripheral BDNF should be treated as an exploratory response marker rather than a validated efficacy biomarker, and BDNF/TrkB-directed interventions should be approached cautiously, as the direction of pathway abnormality may differ across ASD subgroups.

Neuroinflammation and microglial dysfunction: Neuroimmune abnormalities and altered microglial states have been implicated in subsets of ASD, although their causal role, developmental timing, and heterogeneity remain incompletely defined[4,66]. Although M1/M2 terminology has historically been used to summarize pro-inflammatory and reparative programs, single-cell studies demonstrate substantial spatial, temporal, and disease-context-dependent heterogeneity in microglial states[67]. Accordingly, a binary M1/M2 framework is insufficient to represent the multidimensional states observed in vivo. Exercise can induce systemic immune changes, including transient increases in muscle-derived IL-6 and alterations in downstream inflammatory mediators[68]. In human experimental studies, IL-6 administration increased circulating IL-1 receptor antagonist and IL-10[69], whereas exercise and IL-6 infusion attenuated endotoxin-induced tumor necrosis factor-α production[70]. These findings support context-dependent systemic immunoregulatory effects of IL-6 but do not demonstrate direct reprogramming of brain microglia or an ASD-specific mechanism. To our knowledge, direct evidence that exercise-induced, muscle-derived IL-6 crosses the blood-brain barrier and reprograms human microglia in ASD is currently lacking. Accordingly, the proposed pathway should be interpreted as an indirect, hypothesis-level model in which exercise-associated peripheral signals may modify systemic immune conditions, thereby influencing neuroimmune communication. The manuscript therefore does not infer a demonstrated M1-to-M2 conversion in ASD. Validation will require ASD-specific studies linking peripheral mediator kinetics to central neuroimmune measures and functional outcomes.

Energy metabolism and mitochondrial dysfunction: Mitochondrial abnormalities have been reported in subsets of individuals with ASD and may affect ATP generation, redox balance, calcium homeostasis, and other processes relevant to neural function[5,71,72]. However, ASD-associated metabolic abnormalities are heterogeneous, and a single “cerebral energy deficit” model should not be assumed to apply across the spectrum.

Exercise can improve systemic insulin sensitivity and glucose regulation[73,74] and can increase circulating lactate during sufficiently intense muscle activity. During exercise accompanied by elevated arterial lactate concentrations, the human brain can increase lactate uptake[75]. Lactate is an exercise-responsive metabolic substrate and signaling molecule rather than a myokine[13,76]. Under some conditions, lactate can contribute to brain energy metabolism and participate in signaling processes relevant to neuronal function[77]. These findings support a potential link between physical activity and brain energy handling but do not establish correction of ASD-associated mitochondrial dysfunction. Nevertheless, lactate should not be presented as a generic bypass for mitochondrial dysfunction. Lactate and pyruvate participate in a reversible reaction catalyzed by lactate dehydrogenase, and the sustained use of lactate as an oxidative fuel depends substantially on downstream mitochondrial oxidative metabolism[78]. Consistent with this limitation, lactate-supported neuronal activity can involve increased oxygen consumption and ATP synthesis via mitochondrial oxidative phosphorylation[79]. Consequently, any compensatory effect is likely to depend on the specific mitochondrial defect, cell type, developmental stage, and capacity for lactate transport and oxidation. The suggestion that lactate specifically rescues energy-deficient GABAergic interneurons in ASD, therefore, remains at the hypothesis level.

Direct human evidence that exercise-derived lactate corrects ASD-associated mitochondrial dysfunction or mediates behavioral improvement is currently insufficient. Future studies could combine standardized exercise challenges with metabolic phenotyping, magnetic resonance spectroscopy, or other measures to link peripheral substrate dynamics to brain metabolism. Until then, improved systemic metabolic health and lactate-related signaling should be considered plausible but unproven contributors to exercise responses in ASD.

HPA axis and stress response: Altered basal cortisol patterns and stress responsivity have been reported in ASD, although findings vary by developmental stage, sampling methodology, stress paradigm, and clinical subgroup[53,54]. Stress-system dysregulation may interact with emotional and behavioral difficulties in some individuals[54]. However, the available evidence does not support a uniform state of HPA-axis hyperactivation across the autism spectrum.

Higher physical activity has been associated with a steeper diurnal cortisol slope in broader populations[80], although effects are not consistent across all indices of HPA-axis function, and the causal direction remains uncertain. Exercise-responsive mediators such as IL-6 and irisin-related signals participate in immune, metabolic, neuroinflammatory, and neurotrophic pathways[60,68,81], providing plausible indirect links to stress-regulatory systems. Proposed routes include changes in systemic inflammatory conditions and in BDNF-related or other central signaling pathways[60,68,81]. However, these links are largely indirect and transdiagnostic, and evidence that a specific muscle-derived myokine directly recalibrates HPA-axis function in ASD is currently lacking.

Exercise studies in ASD have reported improvements in selected emotional and behavioral outcomes, including those following a structured physical exercise intervention in children with ASD[82], whereas broader syntheses suggest benefits for some social outcomes[6]. These findings do not establish HPA-axis mediation. The observed outcomes are compatible with multiple, nonexclusive mechanisms, including behavioral engagement, changes in sleep or physical fitness, psychosocial effects, and neurobiological adaptation. Therefore, the proposed muscle-HPA link should be treated as a testable hypothesis rather than as an established mechanism. Future trials should combine repeated cortisol measurements with standardized stress paradigms, time-resolved assessments of candidate peripheral mediators, and formal mediation analyses before attributing behavioral improvement to myokine-dependent HPA-axis recalibration.

Linking hypotheses and evidence: Perspectives from preclinical and clinical research

Within the proposed muscle-brain axis framework, it is essential to distinguish evidence that tests mechanisms in ASD from evidence that establishes pathway plausibility in other conditions. The following synthesis, therefore, separates ASD-specific intervention evidence from transdiagnostic mechanistic studies. The latter can identify pathway dependence or sufficiency within their own models but cannot confirm causal mediation in ASD. Table 2 summarizes this evidence and its major translational limitations.

Table 2 Autism spectrum disorder-specific intervention evidence and transdiagnostic mechanistic evidence relevant to the proposed muscle-brain axis.

Evidence tier	Model/population	Intervention/manipulation	Main findings	Relevance to the proposed muscle-brain axis and major limitation	Ref.
ASD-specific exercise evidence	BTBR mice (ASD-relevant model)	5 weeks of voluntary wheel running	Attenuated nociceptive abnormalities; no significant improvement in social interaction deficits	Shows outcome-specific effects of physical activity in an ASD-relevant model. Does not establish myokine mediation or broad behavioral efficacy	[83]
ASD-specific multimodal contextual evidence	Juvenile BTBR mice	Semi-natural housing/environmental enrichment with social and physical components	Improved social behavior and reduced repetitive exploratory behavior	Relevant to multimodal interventions, but not an isolated exercise manipulation; effects cannot be attributed specifically to physical activity	[84]
ASD-specific exercise evidence	Male and female rats with VPA-induced autism-like phenotype	Low- and moderate-intensity interval training	Improved selected anxiety-like and social outcomes; spatial learning and memory did not significantly improve	Supports outcome-specific rather than uniform efficacy. Myokine mediation was not demonstrated	[85]
ASD-specific clinical intervention evidence	Children with ASD	Aquatic exercise intervention	Improved selected executive-function outcomes; serum BDNF increased concurrently	Provides preliminary concurrent human evidence. Co-occurrence does not establish BDNF mediation, skeletal muscle origin, or central target engagement	[62]
ASD-specific clinical synthesis	Autistic children, adolescents, and broader autistic samples across exercise trials	Multiple exercise modalities	Syntheses report benefits for selected social, cognitive, sleep, anxiety, and functional outcomes, but effects are heterogeneous, and some domains show null or inconsistent results	Supports continued clinical investigation while arguing against broad efficacy claims or inference of a single biological mechanism	[6,87-88,102]
ASD-specific mechanistic synthesis	Autistic individuals/ASD-relevant literature	Mechanistic review of exercise interventions	Summarizes candidate neuronal, glial, and gut microbiota-related pathways	Useful for hypothesis generation; mechanistic review evidence does not demonstrate direct causal mediation by myokines	[89,103]
ASD-specific biomarker evidence	Children with ASD	Cross-sectional serum BDNF assessment	Elevated or altered peripheral BDNF has been reported; some studies associate higher levels with symptom severity	Central to the BDNF paradox: Higher peripheral BDNF is not equivalent to beneficial central signaling. Cross-sectional data cannot establish cause, compensation, or tissue source	[57,59]
ASD-specific biomarker evidence	Children with ASD vs healthy controls	Cross-sectional serum irisin assessment	Lower circulating irisin was reported in ASD	Suggests a potential association with physical inactivity or metabolic phenotype but does not establish a causal deficiency or muscle-to-brain mediation	[86]
ASD-specific synaptic mechanism	Mdga2-deficient mice with autism-relevant phenotype	Genetic disruption of a synaptic regulator	Aberrant/excessive BDNF-TrkB signaling was linked to synaptic and autism-relevant behavioral changes	Demonstrates that greater TrkB signaling is not uniformly beneficial and supports caution regarding BDNF/TrkB-targeted strategies. This was not an exercise study	[58]
Transdiagnostic foundational mechanism	Non-ASD mice	Exercise-induced PGC-1α/FNDC5 pathway activation	Exercise-induced hippocampal BDNF through a PGC-1α/FNDC5-related pathway	Provides foundational biological plausibility for an exercise-responsive FNDC5/BDNF pathway. Does not establish ASD-specific mediation	[11]
Transdiagnostic mechanistic evidence	Non-ASD mice; Alzheimer’s disease model	Whole-body FNDC5 loss of function; experimental elevation of circulating irisin	Loss of FNDC5 abolished selected exercise-related cognitive effects; elevated circulating irisin improved selected cognitive and neuropathological outcomes in an Alzheimer’s disease model	Supports pathway dependence or sufficiency within those models. Neurodegenerative findings cannot be treated as causal evidence in ASD	[42]
Transdiagnostic peripheral mechanism	Aging-associated sarcopenia mouse model	Muscle-specific FNDC5 deletion during aerobic exercise	Deletion impaired exercise-related benefits in muscle function and mass	Supports a role for muscle-derived FNDC5/irisin in peripheral exercise adaptation. Does not establish brain or ASD mediation	[50]
Transdiagnostic mechanistic evidence	Non-ASD mouse model	Irisin-neutralizing antibody during exercise	Attenuated selected cognitive and mood-related effects and reduced exercise-associated hippocampal BDNF upregulation and cell proliferation	Supports pathway dependence in a non-ASD setting. Generalization to ASD requires direct validation	[51]
Transdiagnostic molecular mimicry evidence	Non-ASD experimental context; quadriceps muscle	Recombinant irisin administration	Induced skeletal muscle proteomic, metabolic, and microstructural changes resembling aspects of exercise adaptation	Supports partial molecular mimicry in muscle. Does not demonstrate CNS effects, behavioral benefits, or ASD relevance on its own	[52]
Non-ASD pediatric biomarker synthesis	Children aged 5-12 years across randomized exercise trials; not restricted to ASD	Exercise interventions with peripheral BDNF measurement	Only two of five included randomized trials reported significant exercise-related increases in BDNF	Demonstrates that BDNF responses are not uniformly positive even in pediatric exercise studies. Not ASD-specific and cannot establish mediation	[90]

Evidence categories reflect study context, not formal risk-of-bias grading.

ASD: Autism spectrum disorder; VPA: Valproic acid; BDNF: Brain-derived neurotrophic factor; TrkB: Tropomyosin receptor kinase B; PGC-1α: Peroxisome proliferator-activated receptor γ coactivator-1α; FNDC5: Fibronectin type III domain-containing protein 5; CNS: Central nervous system.

Evidence from animal models: Evidence from ASD models suggests that physical activity can affect selected outcomes, but the effects are not uniform. In BTBR mice, five weeks of voluntary wheel running attenuated nociceptive abnormalities but did not significantly improve social interaction deficits[83]. Separately, semi-natural housing that provided environmental enrichment improved social behavior and reduced repetitive exploratory behavior in juvenile BTBR mice[84]. Because this intervention incorporated multiple environmental and social components and was not an isolated exercise manipulation, it is considered here as multimodal contextual evidence rather than exercise-specific evidence. In the valproic acid-induced ASD model, low- and moderate-intensity interval training improved selected anxiety-like and social outcomes but did not significantly improve spatial learning and memory[85]. These findings support outcome-specific effects rather than broad efficacy across ASD-like phenotypes.

Mechanistic gain- and loss-of-function studies provide biological plausibility for FNDC5/irisin signaling, but the available evidence is predominantly non-ASD-specific[42,50-52]. Whole-body FNDC5 deletion has been reported to abolish selected exercise-related cognitive effects in mice[42], whereas muscle-specific FNDC5 deletion in an aging-associated sarcopenia model impaired exercise-related benefits in muscle function and mass[50]. Irisin-neutralization experiments have linked the pathway to selected cognitive and mood-related outcomes, hippocampal BDNF upregulation, and cell proliferation in non-ASD settings[51]. Conversely, experimentally increasing circulating irisin in an Alzheimer’s disease model improved selected cognitive and neuropathological outcomes[42], whereas recombinant irisin administration in another non-ASD experimental context induced skeletal muscle molecular and metabolic changes resembling aspects of exercise adaptation[52].

These studies establish pathway dependence or sufficiency within their respective experimental models; they do not establish that muscle-derived irisin causally mediates exercise benefits in ASD. Neurodegenerative, aging-associated, and neurodevelopmental disorders differ substantially in their pathophysiological contexts. Therefore, these findings are treated here as transdiagnostic mechanistic evidence that generates ASD-relevant hypotheses requiring direct validation in ASD-specific models.

Evidence from human studies: Although animal studies support the muscle-brain axis hypothesis, its clinical relevance must ultimately be validated in humans. Current human studies of individuals with ASD consist mainly of cross-sectional and interventional investigations. These studies provide preliminary support for a potential role of myokines in ASD pathophysiology and exercise responses but also demonstrate considerable complexity and inconsistency.

Cross-sectional studies have examined baseline circulating myokine levels in individuals with ASD. Regarding BDNF, some evidence indicates that serum BDNF levels are markedly higher in children with ASD than in typically developing children and that the magnitude of the increase is significantly associated with Childhood Autism Rating Scale scores. This association identifies peripheral BDNF as a candidate marker for further investigation. However, it does not establish it as a biomarker of disease severity because cross-sectional associations cannot determine causality, central pathway status, or individual predictive validity[59]. This finding is consistent with results from several animal models in which aberrant activation of the BDNF/TrkB pathway induces synaptic dysregulation, suggesting a potential excess rather than a deficiency in neurotrophic signaling in specific ASD subgroups. Evidence regarding the emerging myokine irisin is also inconsistent. Notably, a case-control study reported significantly lower serum irisin levels in children with ASD than in healthy controls[86]. This finding may suggest reduced skeletal muscle endocrine activity or physical inactivity in ASD and raises the possibility that low irisin levels are associated with both metabolic abnormalities and neurodevelopmental dysregulation in ASD.

Interventional studies have examined the potential effects of exercise on behavioral and biological outcomes in ASD. According to a systematic review, physical exercise can improve social communication, executive function, and sleep behaviors in young individuals with ASD[87]. More recent syntheses have extended this evidence base. A meta-analysis of randomized controlled trials reported improvements in sleep and anxiety outcomes in children with ASD, although substantial heterogeneity was observed in the anxiety outcomes[88]. A recent mechanistic review further summarized candidate neuronal, glial, and gut microbiota-related pathways through which exercise may influence ASD-relevant outcomes[89]. Together, these reviews support continued investigation while highlighting that evidence of behavioral efficacy should not be conflated with direct demonstration of the underlying biological mechanisms.

More direct mechanistic evidence comes from studies that include biological measurements. For example, a randomized controlled trial indicated that a 12-week aquatic aerobic exercise regimen improved inhibitory control and cognitive flexibility in children with ASD and increased serum BDNF levels. No comparable changes were observed in the control group[62]. This study shows that peripheral BDNF changes and selected improvements in executive function can co-occur after exercise in children with ASD[62]. However, co-occurrence does not establish that BDNF mediated the cognitive effect, nor does serum BDNF identify a skeletal muscle source or demonstrate central target engagement. The finding should therefore be regarded as preliminary associative evidence warranting further mediation-focused trials.

Positive findings should be interpreted alongside null and heterogeneous results. An integrative review of clinical trial evidence in autistic populations reported no significant overall effect of physical activity on broad autism symptom severity across four trials, no consistent benefit for stereotyped behaviors in a two-trial meta-analysis, and insufficient or inconsistent evidence for some motor outcomes[87]. Biomarker responses are also not uniformly positive. In a 2025 systematic review of five randomized controlled trials in children aged 5-12 years, not restricted to ASD, only two studies reported significant exercise-related increases in BDNF[90]. These discrepancies indicate that behavioral and biomarker effects are likely to depend on the outcome domain, intervention characteristics, and participant population, suggesting that a universal myokine response to exercise is unlikely. The predominance of small studies and positive findings also warrants consideration of selective reporting and possible publication bias. However, these biases cannot be inferred from the present evidence alone.

Despite its promise, the evidence in this field is highly inconsistent for several reasons. First, ASD is highly heterogeneous in terms of genetics, symptoms, and co-occurring conditions. Studies may include subgroups that differ in age, intellectual ability, and other characteristics, leading to substantially different baseline myokine profiles and exercise responses. Second, exercise protocols vary considerably in modality (aerobic, resistance, or combined), intensity, frequency, and duration. Different exercise regimens also produce distinct patterns and rates of myokine release, and no consensus exists regarding the optimal exercise plan for ASD. Third, myokine detection methods, particularly those for irisin, lack standardization. Sampling conditions, including fasting status, circadian timing, and sample processing, can substantially affect measurements and impede direct comparisons among studies. Fourth, cross-sectional designs are inherently limited in their ability to establish causality. Circulating factors such as BDNF are produced by multiple tissues, including platelets, the vascular endothelium, and the brain. Whether changes in their levels are causal, consequential, or compensatory in ASD remains unclear. Although interventional evidence can demonstrate associations, it cannot, by itself, establish that behavioral improvements are directly mediated by specific myokine alterations; such causal questions require dedicated mechanistic designs.

Overall, human evidence is compatible with, but does not establish, the involvement of muscle-brain signaling in ASD. Exercise can alter selected peripheral biomarkers and functional outcomes in some studies, whereas other outcomes are null or heterogeneous. Future research should therefore prioritize adequately powered randomized trials, standardized exercise dosing and biomarker sampling, prespecified subgroup analyses, and formal mediation models. Neuroimaging and metabolic measures may help determine whether peripheral changes correspond to central target engagement. Until such evidence is available, observed biomarker changes should not be interpreted as proof that myokines mediate behavioral improvement.

Translational medicine perspectives: From molecular mechanisms to precision exercise interventions

The translational significance of the muscle-brain axis currently lies primarily in generating testable research strategies rather than validated clinical algorithms. The following sections, therefore, distinguish currently feasible exercise-planning inputs from emerging biomarker approaches and research-only molecular targets. Proposed multimodal combinations are likewise presented as hypotheses requiring prospective evaluation. Figure 2 distinguishes currently feasible inputs from emerging and research-only components. Behavioral acceptability and standard clinical assessment can be implemented now, whereas genotype-guided dosing, serial myokine profiling, and pathway-targeted treatment remain investigational. The figure, therefore, represents a staged research roadmap rather than a validated clinical algorithm.

Figure 2 Hypothesis-generating translational framework for staged exercise research in autism spectrum disorder. Phase 1 distinguishes currently feasible clinical and behavioral inputs from emerging or research-only stratification measures. Phase 2 depicts proposed exercise-responsive muscle and systemic pathways, with biomarkers labeled as candidate research measures and recombinant or small-molecule approaches as investigational targets. Phase 3 shows potential combinations with behavioral therapy, neuromodulation, and pharmacotherapy; direct synergistic efficacy remains to be established. IL: Interleukin; BDNF: Brain-derived neurotrophic factor; TrkB: Tropomyosin receptor kinase B; PGC-1α: Peroxisome proliferator-activated receptor γ coactivator-1α; HPA: Hypothalamic-pituitary-adrenal; ABA: Applied behavior analysis; CBT: Cognitive behavioral therapy; PMC: Premotor cortex; HPC: Hippocampus; PFC: Prefrontal cortex; FNDC5: Fibronectin type III domain-containing protein 5; STAT3: Signal transducer and activator of transcription 3; AMPK: AMP-activated protein kinase; MCTs: Monocarboxylate transporters; Akt: Protein kinase B; IGF-1: Insulin-like growth factor 1; PI3K: Phosphoinositide 3-kinase; CRH: Corticotropin-releasing hormone; ACTH: Adrenocorticotropic hormone; rTMS: Repetitive transcranial magnetic stimulation; tDCS: Transcranial direct current stimulation.

Toward stratified exercise intervention: A feasibility-aware framework: ASD is heterogeneous in its clinical presentation and biological characteristics, including variation in peripheral neurotrophic, inflammatory, metabolic, and stress-related measures[53,54,57,59,91-93]. However, current evidence does not support genotype-guided or myokine-guided exercise prescription in routine clinical practice. A more realistic translational strategy is a staged framework that distinguishes currently feasible clinical inputs from emerging adjunctive measures and research-only stratification approaches.

Exercise planning can incorporate medical safety, baseline physical activity, motor function, sleep, medication use, sensory and behavioral tolerability, social preferences, environmental structure, and previous participation experience. Qualitative studies indicate that participation in physical activity among autistic adolescents depends on interactions among individual characteristics, motivation, social support, and environmental circumstances[94]. Similarly, interviews with autistic youth with mild intellectual disability and caregiver reports identified intrapersonal and interpersonal barriers and facilitators to physical activity participation[95]. These findings support selecting exercise modalities according to tolerability and participation context rather than assuming that a single exercise format is appropriate for all ASD populations. The available literature on exercise program design in ASD can further inform modality selection and implementation considerations[96]. However, it does not establish a molecularly guided prescription algorithm.

Standard clinical metabolic measures and, in research-capable settings, selected inflammatory or cortisol-related measures may help characterize biological heterogeneity[53,54,91,93]. For example, ASD cohorts have shown variability in inflammatory and metabolic profiles, insulin resistance, and mitochondrial abnormalities[91-93]. However, these associations do not provide validated ASD-specific thresholds for matching individuals to exercise modalities, intensities, or doses. Such measures should therefore be treated as descriptive or exploratory adjuncts rather than as determinants of clinical exercise decisions.

Genotyping, multi-omics profiling, serial myokine measurement, and algorithmic decision support should currently be regarded as investigational components of exercise research. Their translation is constrained by assay variability, accessibility, cost, uncertain incremental clinical utility, and the absence of prospectively validated ASD-specific decision rules. Importantly, alterations in BDNF, precursor BDNF, IGF-1, or related biological pathways in ASD[57,59,97] do not establish that genotype or biomarker profiles can be used to select exercise type or dose. To our knowledge, no validated ASD-specific algorithm has demonstrated that genotype- or myokine-guided exercise prescription improves outcomes beyond standard clinical and behavioral assessment. A pragmatic workflow is therefore to screen for medical safety and participation barriers; characterize baseline physical and functional status; select a behaviorally and environmentally acceptable modality; begin with a standardized and tolerable dose; monitor adherence, adverse events, and prespecified functional outcomes; and adjust exercise type, intensity, frequency, or duration according to tolerability and observed response. Biomarkers should initially be embedded as exploratory trial measures rather than used to direct care. This staged framework preserves the long-term goal of stratification while acknowledging current feasibility gaps and avoiding premature claims of precision medicine.

Candidate biomarkers and investigational therapeutic targets: Exercise-responsive mediators are of translational interest, but current evidence is insufficient to classify irisin, BDNF, or inflammatory cytokines as validated response biomarkers in ASD. Circulating concentrations cannot be assumed to provide direct readouts of skeletal muscle secretion or central target engagement. For BDNF, peripheral measurements are substantially influenced by platelet storage and release dynamics[63] and by pre-analytical conditions, including anticoagulant choice, storage time, and temperature[64]. Irisin measurement also requires careful methodological interpretation; circulating human irisin has been quantified by tandem mass spectrometry, highlighting the importance of assay selection when comparing studies[98]. The BDNF paradox is particularly important: Elevated baseline peripheral BDNF in some ASD cohorts[59] cannot be reconciled with the universal assumption that further BDNF elevation is beneficial. Likewise, the concurrent increase in serum BDNF and improvements in selected executive outcomes after exercise in children with ASD[62] demonstrate co-occurrence rather than mediation. The underlying ASD exercise study directly examined a 12-week aquatic intervention, executive outcomes, and BDNF levels. However, such parallel changes do not, by themselves, establish a causal biomarker pathway.

Future trials may nevertheless evaluate candidate biomarker panels prospectively. Such studies should prespecify the sampling time, fasting status, the interval between exercise and blood collection, sample type, processing procedures, and assay platform; distinguish serum from plasma where relevant; measure baseline and longitudinal trajectories; and test whether biomarker changes mediate functional outcomes rather than merely accompany them. Baseline status should also be examined as a potential effect modifier, particularly given heterogeneous BDNF findings in ASD. Until these findings are prospectively replicated, these measures should not be used to predict individual efficacy, monitor adherence, or dynamically adjust exercise prescriptions in routine clinical practice. This caution is supported by direct evidence that peripheral BDNF measurements vary with sample handling and that platelet biology materially influences measured concentrations. Recombinant irisin and downstream pathway modulators, including TrkB-directed compounds, should likewise be framed as investigational. Evidence for recombinant irisin is largely derived from non-ASD experimental models[15,16], and direct extrapolation to ASD is not justified. TrkB-directed pharmacological strategies are of mechanistic interest[14]. However, particular caution is required because an ASD-relevant mouse model has shown that excessive BDNF/TrkB activation can contribute to synaptic and autism-relevant behavioral abnormalities[58]. Any pathway-targeted strategy would therefore require ASD-specific efficacy and safety studies, pharmacokinetic and pharmacodynamic characterization, dose-response evaluation, and stratification by baseline pathway status before clinical testing.

Accordingly, myokines and related pathways are currently best considered tools for hypothesis generation and experimental stratification. Their eventual clinical value will depend on assay standardization, prospective validation, reproducibility across ASD subgroups, demonstration of central or clinically meaningful target engagement, and evidence that biomarker-informed decisions improve outcomes beyond standard clinical and behavioral assessment.

Integrated intervention strategies: Exercise may be most appropriately considered one component of multimodal care rather than a replacement for established behavioral or clinical interventions. Exercise interventions have shown benefits for selected behavioral, cognitive, social, and physical outcomes in ASD, although effects vary across outcome domains and intervention protocols[6]. Behavioral interventions target complementary functional domains; for example, cognitive-behavioral therapy has demonstrated benefits for selected social outcomes in children and adolescents with ASD[99]. However, to our knowledge, direct evidence that myokine-mediated biological changes enhance the efficacy of behavioral therapy in ASD is currently lacking. Such synergy should therefore be tested rather than assumed.

Environmental enrichment provides a useful example of why intervention components must be distinguished. In juvenile BTBR mice, semi-natural housing that provided environmental enrichment improved selected social and repetitive exploratory behaviors[84]. Because this intervention incorporated multiple environmental and social components rather than isolated exercise, its effects cannot be attributed specifically to physical activity and are better interpreted as multimodal contextual evidence.

Similar caution applies to neuromodulation and pharmacotherapy. Brain stimulation approaches, including repetitive transcranial magnetic stimulation, are under investigation for selected ASD-related functional targets[100], whereas antidepressant and anti-anxiety medications have been studied for co-occurring symptoms in autistic populations[101]. However, to our knowledge, evidence for specific exercise-neuromodulation or exercise-drug synergies mediated by muscle-derived signals remains insufficient. Findings from depression, stroke, or other non-ASD conditions should not be treated as ASD-specific evidence of multimodal synergy.

Future trials should therefore evaluate predefined intervention combinations, sequencing, dose, safety, adherence, and prespecified functional outcomes. The multimodal framework is best viewed as a research strategy for testing complementarity among behavioral, physiological, and circuit-level interventions rather than as a currently validated integrated treatment algorithm.

FUTURE DIRECTIONS, CHALLENGES, AND CONCLUSIONS

Unanswered questions and major research gaps

Several questions should be prioritized. First, the developmental trajectory of muscle-brain signaling in ASD is unknown. Longitudinal studies are required to determine whether exercise-responsive mediator profiles change across childhood, adolescence, and adulthood, and whether these trajectories are associated with functional outcomes. Second, ASD-specific causal evidence is needed. Future animal studies should use genetically and etiologically diverse ASD models and, where appropriate, tissue-specific or inducible manipulation of candidate pathways such as FNDC5/irisin signaling. The key question is not whether a pathway mediates exercise effects in aging or neurodegeneration but whether altering a defined muscle-derived signal changes brain and behavioral outcomes in an ASD-relevant model. Third, human studies require greater standardization. Exercise mode, intensity, frequency, duration, adherence, sampling time, fasting status, sample processing, and assay platform should be prespecified. Trials should report both null and positive results and should test mediation rather than infer mechanisms from parallel biomarker and behavioral changes. Fourth, stratified randomized trials are needed. Candidate strata could include baseline metabolic status, inflammatory phenotype, stress-response patterns, or exploratory biomarker profiles, but thresholds must be prospectively validated. Multi-omics and neuroimaging may be useful in mechanistic substudies, although their cost and uncertain incremental clinical value should be acknowledged. These priorities would allow the field to progress from broad biological plausibility toward falsifiable ASD-specific mechanisms.

Challenges

Translation faces both participation and measurement barriers. Sensory sensitivities, motor coordination difficulties, anxiety, environmental intolerance, and variable motivation can affect exercise initiation and adherence. Exercise programs, therefore, require adaptation to individual tolerability and often benefit from collaboration among exercise specialists, occupational therapists, clinicians, caregivers, and behavioral professionals. Methodological limitations are equally important. Myokine assays, particularly those for low-abundance or method-sensitive analytes, are not uniformly standardized, and results can vary according to sampling time, fasting status, circadian rhythm, storage, and processing. Circulating factors also derive from multiple tissues, making it difficult to infer a skeletal muscle origin. Exercise dose is difficult to quantify when adherence and effort vary, whereas diet, sleep, medication use, and baseline physical activity can confound biomarker responses. Finally, many current studies are small, short-term, and heterogeneous. The publication of predominantly positive results may exaggerate apparent consistency. Prospective registration, prespecified outcomes, transparent reporting of null findings, standardized protocols, and data sharing are therefore essential. These limitations currently preclude the use of myokine profiles as validated clinical decision tools.

CONCLUSION

This review supports the muscle-brain axis as an integrative, hypothesis-generating framework for investigating exercise-related biology in ASD rather than as an established treatment mechanism. Exercise-responsive mediators may intersect with pathways relevant to synaptic plasticity, neuroimmune regulation, energy metabolism, and stress responsivity, but the evidential strength varies substantially across mechanisms. ASD-specific causal evidence is limited, and gain- or loss-of-function findings from Alzheimer’s disease, aging, or other non-ASD models provide biological plausibility rather than proof of mediation in ASD. Human studies are preliminary and heterogeneous, and the BDNF paradox further argues against assuming that higher circulating levels of neurotrophic factors are uniformly beneficial. Likewise, direct regulation of human ASD microglia by muscle-derived IL-6 and myokine-mediated recalibration of the HPA axis remain unproven. Accordingly, biomarker-guided exercise prescriptions, recombinant irisin, and TrkB-targeted agents should remain investigational. Progress will require ASD-specific mechanistic studies, standardized biomarker assays, transparent reporting of null findings, and adequately powered, stratified randomized trials that test mediation and clinical outcomes.