Compartment-specific redox modulation in migraine with aura: a systematic review of context-dependent mechanisms

2.1 Study registration and protocol

This review was conducted in accordance with the PRISMA 2020 statement (Supplementary material 1) [20]. We developed a protocol specifying the objectives, search strategy, inclusion criteria, and analysis plan a priori and registered it with the International Prospective Register of Systematic Reviews (PROSPERO; CRD420251143073).

2.2 Eligibility criteria

We included animal studies investigating CSD events relevant to MA and their association with redox pathways without species or age restrictions; sex was prespecified as a sensitivity variable to address potential confounding effects on ROS levels [21]. Eligible studies met the following two criteria: (ⅰ) used any method to induce CSD and (ⅱ) evaluated oxidative stress (OS) markers. OS was defined according to Sies [9] as an alteration in redox homeostasis encompassing both physiological signaling (oxidative eustress) and pathological damage (oxidative distress). Reviews, conference abstracts, comments, duplicate data, and studies that did not comply with animal welfare guidelines were excluded. Specifically, studies that did not meet the Animal Research: Reporting of In Vivo Experiments (ARRIVE) 2.0 reporting guidelines were excluded from the synthesis.

2.3 Literature search strategy

A systematic search was conducted across five databases (MEDLINE via PubMed, Embase, Web of Science, CNKI, and Wanfang) from database inception through June 2025. We combined medical subject headings (MeSH) and free-text keywords to capture four domains: CSD, OS, MA, and rodent models. Citations were deduplicated using EndNote 21.0.1 (Clarivate Analytics, Philadelphia, PA, USA) (Supplementary material 2). Additional unpublished studies were identified through manual reference searches and expert consultation.

2.4 Study selection and data extraction

Two reviewers (Li and Duan) independently screened the titles and abstracts, assessed the full texts, and extracted data using a standardized form (Supplementary material 3). Data were preferentially extracted from the Results section and included (ⅰ) study characteristics; (ⅱ) CSD induction protocols; (ⅲ) redox parameters (categorized into four functional groups, as detailed in the Redox parameter synthesis section) with quantitative values; (ⅳ) bioenergetic indicators, including nicotinamide adenine dinucleotide (phosphate) (NAD(P)H), flavin adenine dinucleotide (FAD), tissue partial pressure of oxygen (pO₂), reported in a single study [14]; and (ⅴ) functional outcomes (e.g., synaptic recovery). Depolarization events were classified according to the terminology used in the original studies (CSD or spreading depression (SD)); both terms refer to the same CSD phenomenon relevant to MA.

2.5 Risk of bias assessment

Two reviewers assessed the risk of bias across ten domains using the SYRCLE tool [22]. Discrepancies were resolved by Kong, and detailed conclusions for all 13 included studies are provided in Supplementary material 4 (Ref. [13, 14, 15, 18, 23, 24, 25, 26, 27, 28, 29, 30, 31]).

2.6 Operational definitions of redox signaling loop components

We defined redox modulation as a change in any of the four parameter groups listed in the Redox parameter synthesis section, thus extending our eligibility criteria for OS to encompass both oxidative shifts and antioxidant responses. We categorized CSD–redox interactions into two functional components based on the following reports in the original study: (ⅰ) forward signaling (CSD-to-redox); i.e., studies reporting statistically significant oxidative shifts (p < 0.05) after CSD induction, as measured using redox-sensitive probes or oxidative damage biomarkers; and (ⅱ) reverse signaling (redox-to-CSD); i.e., studies reporting that redox modulation resulted in significant changes in CSD initiation or susceptibility through triggering, mediating, or modulating mechanisms. No quantitative thresholds were prespecified because of the heterogeneity across studies and the limited standardization of redox parameters.

2.7 Data synthesis and analysis

The heterogeneity across models, induction protocols, and outcomes precluded meta-analysis. We thus conducted a qualitative synthesis of empirically reported CSD–redox interactions and evaluated heterogeneity based on the variations in study design. The synthesis focused on the temporal sequence of redox shifts, the effects of antioxidant interventions, and calcium ion (Ca²⁺) dependency (bioenergetic data were available for one study only [14]). No predefined mechanistic hypothesis was imposed on the data. Subgroup analyses were prespecified according to the experimental model (in vivo vs. in vitro) and species to explore the biological and methodological drivers of heterogeneity. We separately analyzed multiple oxidative endpoints from the same cohort, thereby avoiding aggregation bias while preserving mechanistic insights.

We performed sensitivity analyses to explore robustness by sequentially excluding studies at a high risk of bias or those from a single research group (Kraig’s team [23, 24, 25, 26, 27, 28, 29]). We did not assess publication bias using funnel plots because the synthesis was qualitative. The use of comprehensive searches across multiple databases and unpublished sources was intended to minimize reporting bias.

2.8 Redox parameter synthesis

We extracted and categorized redox parameters from the Results section into four functional groups: (ⅰ) reactive species, such as superoxide anion (O₂⁻) and hydrogen peroxide (H₂O₂); (ⅱ) real-time redox indicators, including recombinant redox-sensitive green fluorescent protein (roGFP) oxidation dynamics, CellROX fluorescence, and NAD(P)H/FAD autofluorescence (available in one study only); (ⅲ) biomarkers of oxidative damage and neuroinflammation, including malondialdehyde (MDA), protein carbonyl (PC) content, and tumor necrosis factor-alpha (TNF-α) protein expression; and (ⅳ) redox-regulatory systems, including superoxide dismutase (SOD) activity, glutathione (GSH) levels, and inducible nitric oxide synthase (iNOS) expression.

2.9 Certainty assessment

We excluded the Grading of Recommendations Assessment, Development and Evaluations (GRADE) [32] framework because it has not been validated for preclinical research. We instead assessed the certainty of the evidence across four domains: SYRCLE risk of bias, robustness across sensitivity analyses, indirectness of evidence to MA in humans, and precision of the outcomes.

3.1 Characteristics of the included studies

Through database screening, 208 records were identified. Thirteen studies (594 animals) met the inclusion criteria [13, 14, 15, 18, 23, 24, 25, 26, 27, 28, 29, 30, 31] (Fig. 1). In eleven studies (85%), only male animals were used [13, 14, 15, 18, 23, 24, 26, 27, 29, 30, 31]. Mixed or unspecified sexes were reported in P0–10 preparations (in vitro) [23, 24, 25, 26, 27, 28]. The rodent models included three species (Table 1, Ref. [13, 14, 15, 18, 23, 24, 25, 26, 27, 28, 29, 30, 31]), with Wistar rats being the predominant model. The studies were published between 2011 and 2024 (Fig. 2). Owing to the substantial methodological heterogeneity across species, induction protocols, and redox detection modalities, we performed a qualitative synthesis of bidirectional modulation patterns rather than a pooled quantitative analysis.

Fig. 1.PRISMA flow diagram of the study selection process. CSD: cortical spreading depression.

Fig. 2.Methodological characteristics of the included studies. Summary of the study characteristics. (A) Publication years of the included studies. (B) Animal species and strains. (C) Methods used for CSD induction. (D) Intervention types and mechanistic targets. roGFP: redox-sensitive green fluorescent protein; CSD: cortical spreading depression; C57BL/6: C57 black 6 inbred mouse strain; KCl: potassium chloride; IL: interleukin; IFN: interferon; IGF: insulin-like growth factor.

3.2 Experimental models and CSD induction parameters

CSD was evoked chemically through focal cortical superfusion with potassium chloride (KCl; 0.5–3 M) in 11 studies [13, 14, 15, 18, 23, 24, 26, 27, 29, 30, 31], whereas electrical pulses were employed to evoke CSD events in 6 studies [23, 24, 25, 26, 27, 28]. The anesthetic agents used varied across the included studies: isoflurane was used in eight studies [13, 15, 23, 24, 26, 27, 29, 31], urethane was employed in two studies [18, 31], ketamine/xylazine was used in one study [14], and pentobarbital was employed in one study [30].

3.3 Risk of bias assessment

The SYRCLE assessment indicated a low risk of bias for seven studies [14, 18, 23, 26, 27, 28, 30] and an unclear risk of bias for six studies [13, 15, 24, 25, 29, 31]. For all 13 studies, the risk of bias associated with random sequence generation and allocation concealment was unclear. The risk associated with blinding of the outcome assessors was considered low risk in one study [29]. Three studies were classified as having a high risk of bias because of reported conflicts of interest [24, 25, 27] (Fig. 3, Ref. [13, 14, 15, 18, 23, 24, 25, 26, 27, 28, 29, 30, 31]).

Fig. 3.Risk of bias assessment. (a) Risk of bias plot. (b) Summary of the risk of bias assessment.

3.4 Redox marker assessment methods

Redox markers were categorized into two distinct detection modalities: real-time kinetic imaging (seven studies [13, 14, 23, 24, 25, 27, 28]) and endpoint biochemical assays (10 studies [15, 18, 23, 24, 25, 26, 27, 29, 30, 31]). Kinetic methodologies provided high temporal resolution through measurements of CellROX fluorescence, roGFP oxidation dynamics, and NAD(P)H/FAD flux. In contrast, endpoint assays primarily quantified MDA levels, SOD activity, PC levels, and H₂O₂ levels. Similarly, six studies relied exclusively on endpoint markers, whereas four studies integrated both modalities to confirm the link between depolarization and OS.

3.5 Mechanistic patterns of compartment-specific redox modulation

This synthesis indicates that compartmentalization determines whether CSD–redox interactions function primarily as a signals or triggers (Table 2, Ref. [13, 14, 15, 18, 23, 24, 25, 26, 27, 28, 29, 30, 31]). We organized the evidence into three interpretive tiers: acute oxidative wavefronts, the efficacy gap between global and compartment-specific modulation, and neuroimmune gating via microglial polarization. Collectively, these tiers suggest a self-reinforcing loop wherein redox shifts modulate cortical excitability.

3.6 Heterogeneity, sensitivity, and subgroup analyses

Biological variability and differences in methodology across animal strains, CSD induction protocols (KCl vs. electrical), and redox detection methods precluded meta-analysis. Excluding studies at a high risk of bias [24, 25, 27] and those originating from a single research laboratory [23, 24, 25, 26, 27, 28, 29] did not alter these findings. This robustness persisted in male-only cohorts [13, 14, 15, 18, 29, 30, 31] after P0–10 rat pups were removed [23, 24, 25, 26, 27, 28]. Further subgroup analyses revealed that an increase in ROS levels following CSD challenge was widespread across rat strains, whereas increases in ROS-mediated CSD initiation appeared to occur primarily in murine preparations [14, 15]. Variations in the anesthetic regimen (isoflurane vs. urethane) and experimental setup (in vivo vs. in vitro) were the primary sources of methodological divergence (Table 1).

4.1 Compartment-specific stabilization of oxidative eustress

IGF-1 appears to exert stabilizing effects through compartment-specific protective mechanisms, selectively attenuating microglial OS without affecting astrocytic ROS levels [25], increasing CA3 neuronal antioxidant capacity [28], or attenuating trigeminal MDA/CGRP release [29]. These findings underscore that maintaining oxidative eustress (physiological redox signaling for adaptive responses) is more critical than indiscriminate ROS elimination alone. Unlike global scavenging strategies, the compartment-specific approach of IGF-1 appears to preserve these signaling gradients while targeting pathological foci, which represents a distinct mechanism and may explain why broad-spectrum interventions often fail to exhibit clinical efficacy.

4.2 Upstream microglial mechanisms and receptor-independent pathways

A systematic review by Anwar et al. [34] indicated that SD triggers neuroinflammation but reported insufficient evidence that the inflammatory burden reciprocally alters SD susceptibility, as the included studies demonstrated either no effect or reduced amplitudes. Given that anatomical constraints prevent the direct diffusion of ROS from the cortex to the trigeminal ganglion, we speculate that CSD-derived ROS may be indirectly associated with peripheral nociception via dural neurovascular intermediaries. Specifically, CSD-induced ROS appear to activate TRPA1 in the cortex and meninges [15], thereby triggering local CGRP release and subsequent neurogenic inflammation, which may eventually sensitize trigeminal nociceptive compartments [31]. Such compartment-specific signaling aligns with recent human fMRI evidence showing altered stimulus-induced hemodynamic responses during MA, which has been attributed to neurovascular uncoupling secondary to CSD [35].

We do not propose TRPA1 as a clinical drug target, as systemic antagonists have repeatedly failed in human trials [36, 37]. Paradoxically, this clinical failure may support our framework by revealing that peripheral blockade cannot interrupt centrally driven, compartmentalized redox-metabolic vulnerability. We utilized TRPA1 as a mechanistic node only to understand how ROS transduce metabolic stress into cortical hyperexcitability [15]. This dissociation between preclinical efficacy and clinical failure underscores the necessity of targeting upstream microglial states (e.g., IGF-1-mediated stabilization) over downstream nociceptive receptors. Intranasal IGF-1 exemplifies how stabilizing local cortical redox signaling can prevent downstream headache generation without direct TRPA1 blockade [29], supporting a shift from global scavenging to compartment-specific modulation.

4.3 Limitations of global scavenging and compartment-specific alternatives

The observed efficacy mismatch suggests that the clinical failure of broad-spectrum antioxidants reflects compartmental disruption [16, 17] rather than a flawed biological hypothesis. Evidence indicates that nonspecific scavenging disrupts oxidative eustress, the physiological redox baseline essential for neuronal function [9], without resolving oxidative distress within localized cellular compartments. This mismatch is particularly evident in studies in which high-dose melatonin was administered [18], where the observed dissociation between CSD kinetics and the levels of static redox markers suggests the collapse of redox homeostasis. The paradoxical acceleration of CSD despite stable systemic MDA or SOD levels [18] challenges simplistic ROS reduction models, thereby refining our understanding of the redox–CSD framework.

Static markers are insufficient for capturing the compartment-specific dynamics governing CSD kinetics. This uncoupling indicates that indiscriminate scavenging likely disrupts physiological redox homeostasis. Such disruption may involve reductive stress [38, 39], an excessive reductive environment that impairs bioenergetic efficiency. Specifically, high-dose melatonin may trigger receptor desensitization or conditional pro-oxidant effects [40], converting neuroprotection into excitotoxicity. Regardless of the underlying mechanism, therapeutic efficacy depends on stabilizing discrete cellular compartments rather than simply reducing the levels of systemic markers. Without real-time metabolic monitoring, these mechanistic interpretations remain speculative.

In contrast, compartment-specific modulation achieves homeostasis without global ROS elimination. Unlike broad-spectrum scavengers, IFN-γ-DC-Exos reduce iNOS expression without altering TNF-α levels [26], indicating that IFN-γ-DC-Exos selectively, rather than completely, suppress inflammation. Similarly, IGF-1-mediated stabilization and microglial M2a polarization [23, 27] may preserve physiological signaling gradients [9, 41] while preventing MC within pathological foci without disrupting oxidative eustress. This dichotomy explains the reported discrepancies: global clearance abrogates adaptive signaling while pathological foci are left intact, whereas targeted interventions maintain redox homeostasis within the range of oxidative eustress. These findings suggest that microglial phenotype switching and selective mitochondrial protection should be prioritized over global scavengers in the development of future strategies.

4.4 Translational feasibility

Our findings have several meaningful clinical implications. First, intranasal IGF-1 attenuates trigeminal CGRP release and the expression of oxidative markers in preclinical models [29], suggesting that a noninvasive delivery route may circumvent the limitations of systemic scavenging strategies. Second, the M2a polarization mechanisms identified here (environmental enrichment paradigms, phasic IFN-γ modulation, and exosomal miR-219 release) [23, 26, 27] are promising in principle, but their pharmacological development for human use remains unrealized. Third, for clinical translation, staged validation is needed, including replication in female rodent models to resolve the 85% male bias [42], integration with noninvasive metabolic monitoring methods (e.g., ³¹P-MRS) [43, 44, 45] to identify patient subgroups with compartment-specific vulnerability, and proof-of-concept trials targeting high-risk windows such as the perimenstrual phases when bioenergetic instability peaks [46, 47].

4.5 Sex-specific translational constraints

The evidence presented herein was derived almost exclusively from male rodents (85% of the included studies) [13, 14, 15, 18, 23, 24, 26, 27, 29, 30, 31], which represents a mismatch considering the approximately 3:1 female-to-male prevalence ratio among individuals with clinical migraine [1]. According to the Sex and Gender Equity in Research (SAGER) guidelines [42], this profound sex bias represents a critical translational blind spot that severely limits clinical generalizability. Compared with male brains, female brains have significantly greater resident microglial density, and estrogen suppresses microglial activation. This increased microglial density and hormonal modulation may narrow the functional margin of safety in females, potentially representing a contributing factor to the female-predominant migraine phenotype; however, this mechanism remains untested in females, among whom the disease burden is greater. Furthermore, the estrogen-dependent efficacy of IGF-1 suggested by fundamental neurotrophin interactions remains untested in migraine models. If this efficacy is reproducible in female cohorts, intranasal IGF-1 could serve as a stratified therapeutic strategy during vulnerable windows, such as perimenstrual periods [46, 47], when bioenergetic instability may most substantially compromise CSD threshold resilience, offering an alternative to global ROS scavenging [16, 17].

4.6 Methodological considerations and validation

The CSD–redox response was characterized by substantial heterogeneity, arising from the interplay between induction modalities and metabolic states [13, 14, 15, 18, 29, 30, 31]. The reversibility of the redox response differed markedly according to the induction mode. KCl-induced roGFP oxidation partially recovered, whereas hypoxia and the mitochondrial uncoupler carbonyl cyanide-4-(trifluoromethoxy)phenylhydrazone (FCCP) caused persistent oxidation. Ames characterized the brain as operating with a razor-thin bioenergetic “margin of safety” [48]. Malkov et al. [14] demonstrated that chemical inhibition of glycolysis converts transient redox shifts into persistent MC. However, because this evidence was derived from a single study using acute slice preparations under extreme artificial bioenergetic stress conditions, generalizing ROS as obligatory triggers of clinical migraine extends beyond the current evidence and requires direct validation.

While seemingly contradictory, methodological variations can be reconciled within a coherent spectrum under a compartmentalization framework. These findings suggest that outcomes depend on specific cellular niches rather than systemic ROS levels, reconciling the distinct efficacies of global and targeted antioxidant strategies. Moreover, the sensitivity of redox measurements depended on the anesthetic regimen used. Isoflurane, which was employed in eight of the thirteen studies (Table 1), substantially reduced susceptibility to CSD compared with urethane, whereas neither anesthetic significantly altered propagation kinetics [49]. These pharmacological discrepancies may have artifactually increased CSD thresholds, potentially masking the potency of pro-oxidant triggers. Furthermore, static oxidative markers (MDA and SOD), despite their inclusion in our four functional parameter groups, failed to capture the transient redox wavefronts that propagate with CSD events.

Additionally, ³¹P-MRS revealed potential interictal phosphorylation deficits in migraineurs [43, 44, 45], which are not consistent with our animal data but are consistent with clinical observations. The chronic bioenergetic instability observed in human patients conceptually aligns with the acute MC reported by Malkov et al. [14], suggesting that mitochondrial stabilization may represent translational target. However, the causal link between chronic deficits and acute failure remains untested. Compounding this uncertainty, the 85% male predominance in our synthesized evidence [13, 14, 15, 18, 29, 30, 31] likely obscures sex-specific metabolic vulnerabilities, particularly those shaped by microglial sexual dimorphism [50].

4.7 Limitations and future directions

We acknowledge three limitations of this synthesis: the 85% male bias, reliance on acute stress models, and substantial methodological heterogeneity. All the studies had an unclear risk of bias with respect to sequence generation and allocation concealment. Experimental CSD does not replicate spontaneous migraine onset or pain [51]; these models primarily capture aspects of aura neurobiology rather than the complete syndrome.

Future research should transition from static endpoint assays (MDA/SOD) to the real-time dynamic monitoring using probes (roGFP/CellROX) [13, 14]. This shift is essential for capturing the rapid kinetics of oxidative wavefronts. Cell-specific mapping must be implemented to distinguish the distinct contributions of neuronal and glial compartments [23], as tissue-level analyses obscure specific cellular control points. As noted in the Results section, clodronate depletion confirms the necessity of microglia [23] but does not establish whether protection reflects redox signaling or physical interactions, underscoring the need for cell-specific genetic approaches. Furthermore, experimental designs must incorporate female cohorts and adhere to the SAGER guidelines [42]. Researchers must control for ovarian hormone fluctuations, as estrogen modulates both microglial reactivity [52] and CSD susceptibility [21]. Finally, the current evidence primarily establishes correlations. Validating the causal chain requires the integration of optogenetic tools with real-time redox probes to selectively manipulate cell populations while monitoring local redox homeostasis. Such technological convergence is necessary to define the point of no return within the CSD feedback cycle.