As the largest and most intricate microbial ecosystem in the human body, the gut microbiota is composed of approximately 3.8 × 10¹³ microorganisms. Its composition and function are shaped by a multitude of factors—including host genetics, dietary patterns, lifestyle, and disease status—resulting in pronounced inter-individual heterogeneity and dynamic shifts across the human lifespan (20, 21). Through the neuroendocrine, immune, and metabolic axes of the gut–brain axis, the gut microbiota establishes a multi-layered, bidirectional communication network with the central nervous system. This network exerts profound effects on brain development, information processing, and the aging process (22, 23).

Dysbiosis refers to a disrupted state of gut microbial composition and function and is closely linked to both cognitive decline and chronic inflammation (40, 41). The underlying mechanisms involve several key pathways.

Phytochemicals are non-essential, bioactive trace compounds widely found in plant-based foods. To date, over 100,000 types have been identified, with the major groups including polyphenols, carotenoids, isothiocyanates, organosulfur compounds, terpenoids, and phytoestrogens (59). Although these compounds are not essential nutrients for humans, they have attracted significant attention due to their antioxidant, anti-inflammatory, and chronic disease prevention properties. Their fate in the human body is determined by factors such as molecular size, lipophilicity, blood–brain barrier (BBB) permeability, chemical stability, and, crucially, gut microbial metabolism (60, 61). Many parent molecules, due to their polarity or steric hindrance, have difficulty crossing the BBB. However, after transformation by microbial-specific enzymes into sulfate or glucuronide conjugates, these metabolites can be detected in brain tissue and often exhibit enhanced neuroactivity (62, 63). The classification, typical dietary sources, and representative bioactivities of major dietary phytochemicals are summarized in Table 1.

Current mechanistic studies mainly focus on: (1) scavenging free radicals and activating the Nrf2–ARE antioxidant pathway (64), (2) inhibiting the NF-κB–MAPK inflammatory cascade (65), (3) modulating neurotransmitter and neurotrophic factor levels (66), and (4) epigenetic reprogramming via histone acetylation and DNA methylation (66). However, systematic comparisons of the pharmacokinetics, dose–response relationships, and BBB permeability among different classes of phytochemicals remain limited (60, 67, 68). This gap has partially constrained the extrapolation of results and their clinical translation.

Early research primarily attributed the neuroprotective effects of phytochemicals to their direct antioxidant and anti-inflammatory actions. However, pharmacokinetic data demonstrate that most parent phytochemical molecules exist in plasma and cerebrospinal fluid at extremely low free concentrations, largely in conjugated forms. This indicates that the true “active agents” mediating neuroprotection are likely microbiota-derived metabolites (63). From a prebiotic perspective—defined here as the ability of dietary components to selectively promote the growth and metabolic activity of beneficial gut microbes—phytochemicals serve dual roles: they act as “metabolic substrates” for the microbiota and simultaneously “reshape the microbial ecosystem” by selectively inhibiting or promoting specific taxa (16, 69) (see Figure 2).

For example, interventions with berry anthocyanins have been shown in both animal and human studies to increase the abundance of butyrate-producing bacteria and elevate short-chain fatty acid (SCFA) levels, with these changes correlating with improved spatial memory (70). Importantly, phytochemicals may exhibit a “double-edged sword” effect—meaning their benefits and risks depend on dose and duration. For instance, oral curcumin has been shown to ameliorate memory deficits in APPswe/PS1dE9 mice by inhibiting the HMGB1–RAGE/TLR4–NF-κB pathway, yet high doses or prolonged use can reduce hepatic and renal iron content and adversely affect reproductive function (71–73).

Recent evidence from a meta-analysis of 24 trials (N = 2,336) indicates that long-term polyphenol supplementation in adults aged ≥60 produces only mild improvements in immediate recall, with no significant effects on delayed recall or executive function and considerable between-study heterogeneity (I² > 60%) (74). Similarly, systematic reviews of curcumin interventions report that only two double-blind RCTs have observed meaningful cognitive benefits in populations with cognitive impairment or dementia; most studies in healthy or metabolically abnormal individuals have shown limited advantage (75). Furthermore, a 12-week matcha intervention in Japanese older adults improved emotion recognition and sleep quality but did not yield statistically significant changes in MMSE or MoCA scores (76).

The mechanistic targets of different phytochemicals vary considerably.

Polyphenols primarily enhance synaptic plasticity and antioxidative defenses (77); isothiocyanates are more associated with detoxification and anti-inflammatory pathways (78); and terpenoids (e.g., curcumin) exhibit both anti-inflammatory and epigenetic regulatory effects (79).

Nonetheless, several key challenges remain: the majority of RCTs have relatively short follow-up periods ( ≤ 16 weeks), highly variable intervention doses, delivery matrices (e.g., food vehicles or supplement forms), and marked differences in baseline microbiota composition. This limits the ability to detect cumulative cognitive effects. There is a lack of integrated multi-omics approaches; only a few studies concurrently assess metagenomic and metabolomic changes, making it difficult to establish a robust causal “phytochemical–microbiota metabolism–central phenotype” chain. Variability in individual microbiome backgrounds further complicates efficacy assessment and clinical translation.

Collectively, current evidence suggests that the cognitive benefits of phytochemicals are largely dependent on their biotransformation into low-molecular-weight metabolites by the gut microbiota and the consequent remodeling of microbial community structure. However, the effectiveness of these interventions is limited by factors such as dose, intervention duration, and inter-individual microbiome variability. There is a clear need for long-term, stratified, multi-omics randomized trials to elucidate optimal compound–microbiota matching strategies for cognitive health and advance the clinical translation of precision diet–microbiota interventions, thereby capturing the full complexity of the diet–microbiota–brain axis.

Pomegranate, strawberries, and walnuts are notable dietary sources of ellagitannins. These compounds are hydrolyzed in the upper gastrointestinal tract to release ellagic acid (EA). In the colon, ellagic acid undergoes sequential dehydroxylation, ring opening, and decarboxylation, catalyzed by gut bacteria such as Gordonibacter and Ellagibacter. This microbial transformation leads to the production of urolithin A (Uro-A) and urolithin B (Uro-B) (92, 93). The presence or absence of the ellagic acid decarboxylase gene cluster defines three metabolic phenotypes: UM-0 (non-producers), UM-A (Uro-A predominant), and UM-B (Uro-B and iso-Uro-A predominant), reflecting inter-individual variability in microbial metabolism (94).

Animal studies have shown that oral Uro-A (300 mg·kg⁻¹ for 14 days) significantly improves spatial memory, reduces neuronal apoptosis, and promotes neurogenesis in APP/PS1 mice (95). Pharmacokinetic data indicate that Uro-A is rapidly absorbed, with peak brain concentrations achieved 4 h after a single dose, and is present mainly as glucuronide, sulfate, and methylated derivatives. Brain tissue concentrations (28–35 ng·g⁻¹) are higher than plasma levels, suggesting selective enrichment (96). Uro-A modulates key anti-aging and autophagy pathways (AMPK, SIRT, mTOR), although its direct molecular targets require further elucidation (97). Urolithin C (Uro-C) has also been shown to reduce Aβ_1–42 deposition and maintain cholinergic balance in aging models (98).

While preclinical evidence robustly supports the neuroprotective and anti-inflammatory roles of urolithins, human data remain limited and inconsistent. Uro-A is generally recognized as safe (GRAS) by the FDA and exhibits moderate BBB permeability and multi-target mitochondrial benefits (99, 100). However, the absence of large, dose-escalation, and long-term RCTs, particularly those stratified by UM phenotype, is a major gap in clinical translation. Overall, urolithins are promising agents for cognitive protection, but future studies should focus on population stratification and rigorous long-term evaluation.

Cruciferous vegetables such as broccoli and kale are rich in glucoraphanin, which is hydrolyzed by plant myrosinase or microbial β-thioglucosidase to produce sulforaphane (SFN). SFN covalently binds cysteine residues in Keap1, leading to Nrf2 nuclear translocation and activation of ARE-dependent antioxidant and detoxification genes. In addition, SFN reversibly inhibits HDAC through thiol alkylation, increasing histone acetylation and upregulating neurotrophic factors such as BDNF (101, 102).

Twelve-week dietary supplementation with SFN in animal models enhances hippocampal PGC-1α, NRF-1, and TFAM transcription, promotes mitochondrial biogenesis, and mitigates age-related cognitive decline (103). SFN also increases BDNF expression and synaptic plasticity through HDAC inhibition in AD mouse and neuron models (102). Early clinical trials are encouraging: a 12-week, double-blind RCT of 30 mg/day SFN improved spatial orientation and working memory in traumatic brain injury patients (104), while similar interventions in older adults improved overall cognitive performance and selectively benefited processing speed and working memory (105, 106).

However, current studies are limited by small sample sizes, lack of dose-gradient design, and short follow-up durations. Long-term safety and the minimum effective dose remain unclear, emphasizing the need for multi-center RCTs with pharmacokinetic monitoring. In summary, SFN demonstrates strong neuroprotective mechanisms across preclinical and early clinical studies, but definitive large-scale evidence is still needed.

Daidzein, a major soy isoflavone, is metabolized by gut bacteria such as Slackia isoflavoniconvertens into equol. Only 30–50% of Asian adults are equol producers, underlining the critical influence of the gut microbiota phenotype on efficacy (107, 108). Equol supplementation activates the ERβ-PI3K–Akt signaling pathway and improves spatial memory in animal models (109).

Epidemiological evidence from large Japanese cohorts supports a link between soy isoflavone intake and reduced cognitive impairment risk (110, 111), though most studies do not directly assess equol producer status. Cross-sectional studies show that S-equol producers exhibit better cognitive scores and lower MCI prevalence (112), but findings in other populations (e.g., Singaporean, Chinese, and Asian-American) are inconsistent, likely due to lower isoflavone intake and lack of phenotype assessment (113, 114).

Meta-analyses of RCTs suggest that soy isoflavones provide small improvements in global cognition and memory, but with considerable limitations in sample size and follow-up (115, 116). Mechanistically, equol combines estrogenic, antioxidant, and anti-inflammatory properties, affecting vascular function, metabolic homeostasis, and neuroinflammation—all relevant to the prevention of vascular cognitive impairment (VCID). Future research should stratify participants by equol producer status, use long-term, dose-gradient RCTs, and comprehensively monitor estrogen-related adverse effects. Overall, equol represents a microbiota-dependent, multi-modal neuroprotective agent with particular potential in precision interventions.

Citrus flavonoids such as hesperidin and naringin are converted by microbial O-methyltransferase–positive bacteria (e.g., Clostridium orbiscindens) into hesperetin and naringenin. These conversions require microbial cleavage of glycosidic bonds and demethylation (117). Alzheimer's disease (AD) features progressive memory decline, Aβ plaque deposition, tau hyperphosphorylation (via GSK-3β activation), and impaired insulin signaling (118–121). In this pathological context, naringenin has shown anti-inflammatory, antioxidant, anti-apoptotic, and neuroprotective effects (122).

In AD models, naringenin improves spatial learning and memory, modulates the PI3K/AKT/GSK-3β pathway, reduces tau phosphorylation, restores insulin signaling and PPAR-γ activity, and confers both metabolic and neuroprotective benefits (123, 124). In vitro, naringenin protects against Aβ-induced apoptosis in neuronal cells by regulating caspase-3, PI3K/AKT, and GSK-3β (125). Animal studies show that oral naringenin reduces hippocampal lipid peroxidation, neuronal apoptosis, and reverses memory loss; its neuroprotection is partly estrogen receptor–dependent (126). Naringenin also inhibits AChE and BACE1, restoring memory, and targets multiple AD-related enzyme pathways (127–129).

Collectively, current evidence supports naringenin's multi-target protective actions against AD-related cognitive impairment via the modulation of amyloid/tau pathology, PI3K/AKT/GSK-3β and insulin pathways, cholinergic neurotransmission, the CRMP2 axis, and enzyme inhibition, as well as through antioxidative and anti-inflammatory mechanisms. Its estrogenic and metabolic regulatory properties further reinforce its candidacy as a multi-target nutraceutical-pharmaceutical for future clinical evaluation.