Parkinson’s disease (PD) is a neurodegenerative disease characterized by resting tremors, slow movement, muscle rigidity, as well as some non-motor symptoms and secondary motor symptoms (Tolosa et al., 2006; Jankovic, 2008). Numerous environmental and genetic factors influence the risk of PD, with different factors predominating in distinct patients. These factors converge on specific pathways, including mitochondrial dysfunction, oxidative stress, protein aggregation, impaired autophagy, and neuroinflammation (Simon et al., 2020). It is the second most prevalent neurodegenerative disease (Ben-Shlomo et al., 2024). In high-income countries, the median age-standardized annual incidence rate of PD is approximately 14 cases per 100,000 individuals (Ascherio and Schwarzschild, 2016). Currently, the diagnosis of PD is based on clinical criteria, with the standard definition encompassing bradykinesia accompanied by resting tremor, rigidity, or both(Bloem et al., 2021). The current treatment for PD primarily involves levodopa to alleviate motor symptoms, supplemented by dopamine agonists or B-type monoamine oxidase inhibitors based on individual patient profiles (Foltynie et al., 2024). For all PD patients, current therapies are symptomatic, focusing on improving motor and non-motor signs and symptoms, with no medications demonstrating definitive evidence of disease-modifying effects (Armstrong and Okun, 2020; Vijiaratnam et al., 2021). However, PD may have a prodromal phase during which precise and early diagnosis based on clinical presentation remains challenging (Bloem et al., 2021). In addition, clinical diagnosis of PD typically achieves an accuracy of only 80–90% when compared to pathological confirmation (Ascherio and Schwarzschild, 2016). This urgency has driven significant interest in the study of biomarkers for PD, highlighting the critical need for more convenient, accurate, and early diagnostic methods. Currently, the primary sample sources for clinical research on biomarkers for PD are blood and cerebrospinal fluid (CSF) (Siderowf et al., 2010; Aasly et al., 2012; Abbasi et al., 2018; Ng et al., 2020; Aamodt et al., 2021; Mao et al., 2023). However, both blood and cerebrospinal fluid collection are invasive procedures, which can impose substantial psychological stress and financial burden on patients.

During the patients’ clinical visits, we observed that the tongue coating (TC)of PD patients exhibited a notable thick and greasy appearance, with some presenting abnormal white or yellow discoloration. This piqued our interest. TC refers to the layer present on the dorsal surface of the tongue, which serves as a critical observation target in the “visual diagnosis” of traditional Chinese medicine (TCM). It is primarily composed of a complex mixture of microorganisms (bacteria), keratinized epithelial cells, saliva, blood metabolites, and food residues (Danser et al., 2003; Van Tornout et al., 2013; Seerangaiyan et al., 2018). The appearance of TC is typically described based on its color and texture. Coating colors are generally classified into four main categories: white, yellow, gray, and black. Texture, on the other hand, encompasses various types, including thin, thick, smooth, dry, greasy, rotten, peeled, and the presence or absence of TC roots. While the study of TC has been extensively explored within the framework of traditional Chinese medicine, recent research has increasingly focused on advanced techniques such as image acquisition, digital processing, and computer-aided analysis (Kim et al., 2013; Segawa et al., 2021; Yuan et al., 2023). Research has suggested that tongue images provide a stable and reliable method for diagnosing gastric cancer, offering significant advantages over traditional blood biomarkers (Yuan et al., 2023). Research on TC is rapidly advancing, with its applications gradually extending beyond oral (Zhang et al., 2024) and gastrointestinal diseases (Cui et al., 2019; Mu et al., 2019; Guo et al., 2022) to encompass a broader spectrum of health conditions (Kapila, 2021; Pathak et al., 2021). Beyond image processing, the study of TC composition has gained increasing attention. TC sampling is straightforward, non-invasive, and low-risk, making it an attractive and promising material for clinical research. Emerging evidence suggests that TC holds significant potential as a non-invasive diagnostic tool for disease detection and as a means to monitor disease progression and prognosis (Ali Mohammed et al., 2021; Li et al., 2021).

Building on the aforementioned insights, we are particularly interested in identifying biomarkers within the TC that could aid in the diagnosis and prognosis of PD. The unique composition of TC presents an untapped potential for uncovering disease-specific metabolic signatures. Ultra-Performance Liquid Chromatography coupled with Quadrupole Time-of-Flight Mass Spectrometry (UPLC-Q/TOF-MS) has been extensively applied in metabolomics research, demonstrating its capability to analyze complex biological samples and elucidate metabolic alterations in various diseases (Cui et al., 2018; Shao et al., 2021; Zhao et al., 2021; Liu et al., 2022; Yang et al., 2022). 16S rRNA (16S ribosomal RNA) sequencing technology is widely used in microbiome research as it provides insights into the composition of microbial communities. It is a reliable method for high-throughput sequencing analysis (Curry et al., 2022; Hassler et al., 2022). Thus, we propose to conduct a non-targeted metabolomics study of TC samples from Parkinson’s patients using UPLC-Q/TOF-MS. At the same time, we applied 16S rRNA sequencing to study the microbiome of PD patients, aiming to identify differences in their microbial communities. By integrating untargeted metabolomics and microbiome analysis, we hope to fill the gap in research in this area. Our goal is to identify reliable biomarkers that could serve as non-invasive diagnostic tools and provide new therapeutic targets, paving the way for innovative approaches in the diagnosis and treatment of PD.

TCM has a long history in treating PD, and modern research in this field is also progressing rapidly. For example, studies have suggested that acupuncture holds significant promise as a therapeutic approach for PD, showing potential benefits in alleviating symptoms and improving patients’ quality of life (Zhuang and Wang, 2000; Zuo et al., 2022). This provides valuable insights into exploring TCM’s potential in diagnosing and treating PD. As a vital part of TCM, TC can provide a micro - level explanation of diseases through component research. Most studies by international scholars on TC have primarily focused on its association with halitosis (Roldán et al., 2005; Liu et al., 2006; Romano et al., 2020). Moreover, it is quite common internationally to use the complete removal of TC as one of the methods for treating halitosis (Kostka et al., 2008). Additionally, some international studies have explored the relationship between TC and systemic diseases beyond the oral and digestive systems (Kim et al., 2013; Mu et al., 2019; Hao et al., 2021; Xu et al., 2021), such as early rheumatoid arthritis (Kroese et al., 2021), pneumonia(Takeshita et al., 2010), lung cancer (Su et al., 2011), chronic renal failure (Gulsahi et al., 2014; Hao et al., 2019), coronary heart disease (Hao et al., 2019), perioperative conditions (Funahara et al., 2018), osteoporosis (Pereira et al., 2018), ischemic stroke (Huang et al., 2022), menstrual pain (Kim et al., 2017) and so on. However, most of these studies focus on the visual characteristics of TC, with limited investigation into its composition. Compared to the superficial visual features of TC, studying its composition offers deeper insights into disease mechanisms at the molecular level. Therefore, in clinical practice, when observing distinctive TC characteristics in outpatient cases, our team places greater emphasis on and interest in research focusing on its composition.

Our experimental results suggested a significant difference in TCS between the PD and CON groups, prompting us to focus on molecular studies of TC composition to further explore the unique characteristics of PD patients. To our knowledge, no studies have yet investigated the microbial composition of TC samples from PD patients. However, studies (Fan et al., 2022; Hu et al., 2024) on fecal samples from PD patients have shown significant increases in the relative abundance of Proteobacteria, Firmicutes, and Actinobacteria at the phylum level, while Bacteroidetes exhibited a significant decrease. At the genus level, Streptococcaceae increased, and Veillonellaceae decreased, with LEfSe analysis showing significant enrichment of Bifidobacterium in PD samples. These findings align with the microbial alterations observed in TC samples from PD patients in our study. Compared to fecal samples, TC sampling is more easily accepted by patients and is also more convenient, clean, and safe for the operators. KEGG pathway analysis in our study suggested that metabolism-related pathways were notably enriched, which may reflect increased metabolic activity within the microbial community. This has sparked further interest in the metabolomics profile of the TC samples. So, we also opted for non-targeted metabolomics due to its high sensitivity, comprehensive nature, and ability to provide an intuitive reflection of biological states. The UPLC-Q/TOF-MS system was selected for its advantages, including high separation efficiency, exceptional mass resolution and sensitivity, rich structural information, and strong stability and reproducibility, making it highly suitable for analyzing a wide range of metabolites in complex biological samples. Given that most PD patients are elderly and often present with underlying conditions such as hypertension and hyperlipidemia, these factors are unavoidable in clinical sample collection. To mitigate their potential influence on our study, we selected family members accompanying patients for follow-up visits, matched by age group, as the CON group. This approach helped reduce the confounding effects of environmental and lifestyle factors. Statistical analysis showed no significant differences in the prevalence of hypertension and hyperlipidemia between the PD and CON groups, further enhancing the reliability of our research. At the same time, we combined image-based evaluation with metabolomics assessment to fully leverage the advantages of TC analysis. Our results suggested the robustness and reliability of our model. In our study, we identified 31 significant differential compounds. Some differential compounds, such as the upregulation of Docosanamide, originate from external contamination, while others, like the downregulation of Palmitoylethanolamide (PEA), are primarily derived from the body itself. This suggests that the disease state of Parkinson’s patients is influenced by both external environmental factors and internal factors. The utility of PEA has been widely studied, particularly in the context of neurodegenerative diseases. The mass spectrometry results of PEA are shown in Figures 4B,C. The Spearman correlation analysis between metabolomics and microbiome data is shown in Figure 8.

The relationship between the oral-gut axis has been extensively studied. Building on this, research on the oral-gut-liver axis has also gained traction, leading to discoveries such as the connection between periodontitis and cardiovascular diseases (Koren et al., 2011; Acharya et al., 2017). This highlights the inseparable link between the oral cavity and the gut while offering promising avenues for exploring the oral-gut axis and its derivative pathways. For example, the oral-gut-brain axis. The gut microbiota plays a key role in communication between the brain and the gut (Aburto and Cryan, 2024). Increasing evidence suggests that impaired gut barrier function is associated with a wide range of central nervous system disorders, including neurodevelopmental, psychiatric, and neurological diseases. This has expanded the perspective of brain-related conditions to systemic diseases, with PD being one of the most extensively studied neurological conditions describing the relationship between the gut barrier and brain barrier (Aburto and Cryan, 2024). Studies have shown that one of the hallmarks of neurodegenerative diseases is alterations in the gut microbiota. PEA plays a role in several physiological processes directly related to maintaining gut barrier function, regulating inflammation and pain, and energy metabolism (Russo et al., 2018). PEA is an endogenous cannabinoid ethanolamide that acts as an “on-demand molecule” in the central nervous system, produced and released from neurons and glial cells. It is known for its analgesic, anti-inflammatory, and neuroprotective properties, achieved through the activation of various receptors and ion channels (Cordaro et al., 2018; Kiani et al., 2020; Cifelli et al., 2022). PEA belongs to the N-acylethanolamine (NAE) family, a group of bioactive lipids capable of modulating peripheral and central pathological processes (Russo et al., 2018). Neuroinflammation plays a crucial role in the pathogenesis of neurodegenerative diseases such as Alzheimer’s disease (AD) and PD (Siracusa et al., 2015; Crupi et al., 2018; Cordaro et al., 2020). PEA, with its potent neuroprotective and anti-inflammatory properties, is a key player in resolving neuroinflammation (Skaper et al., 2015; Cordaro et al., 2020; Petrosino and Schiano Moriello, 2020). Its primary target is the nuclear peroxisome proliferator-activated receptor-α (PPAR-α). Additionally, PEA can activate and regulate transient receptor potential vanilloid 1 (TRPV1) channels and indirectly activate cannabinoid receptors CB1 and CB2 (Esposito et al., 2012; Petrosino and Schiano Moriello, 2020; Cifelli et al., 2022). Through the CB2-mediated anti-inflammatory pathway, PEA modulates microglial polarization, reduces the release of pro-inflammatory cytokines, and enhances migration and phagocytic activity (Cifelli et al., 2022). By binding to GPR55 receptors, PEA enhances GABAergic transmission in the striatum and increases the postsynaptic synthesis of endogenous cannabinoid 2-AG, thereby regulating GABA release via presynaptic CB1Rs (Cifelli et al., 2022). This multifaceted mechanism endows PEA with exciting potential in treating PD and controlling its progression, as supported by various studies (Siracusa et al., 2015; Avagliano et al., 2016; Brotini et al., 2017; Crupi et al., 2018; Brotini, 2021; Palese et al., 2022). Moreover, PEA has shown promising effects in improving cognitive decline and mild cognitive impairment (MCI), particularly in the early stages of cognitive disorders (Calabrò et al., 2016; Colizzi et al., 2022; Landolfo et al., 2022). In our results, the PEA level in the PD group is significantly downregulated, making it an important factor distinguishing the PD group from the CON group. This further supports previous affirmations regarding the effects of PEA. Our findings indicate that the deficiency of PEA is a significant characteristic of PD patients and may even be a potential underlying cause of PD. This result is also evident at the level of TC, suggesting the potential for further research. It suggests that TC plays an important role in differentiating between individuals with or without PD. Additionally, since TC is a convenient and non-invasive sample type, it emphasizes its advantages and potential. Therefore, our study indicates that TC could serve as a promising sample for distinguishing PD, with PEA as a key biomarker for this differentiation. Given that PEA is endogenous, it is a more accurate reflection of the host’s condition. We have summarized the relevant routes of PEA based on literature, as shown in Figure 9 (Costa et al., 2008; Ueda et al., 2013; Aguilera et al., 2015; Musella et al., 2017; Lee et al., 2020; Mirzaei et al., 2021; Cifelli et al., 2022; Cong et al., 2022; Ikeda et al., 2022).

In our study, PEA appeared to play a significant role in differentiating between the PD and CON groups, but was not found to be significantly correlated with cognitive (based on MMSE scores) or disease progression (based on H-Y staging) groupings. This is because the difference in PEA levels between the CON and PD groups is substantial. In the CON group, the mass spectrometry peak value of PEA was relatively high, reaching hundreds of thousands, while in the PD group, the relative peak value was only 1. Therefore, PEA could be clearly separated between these two groups, where it serves an important role. However, when the relative peak value of PEA is uniformly 1 in the PD group across all samples, it cannot be used as a marker for cognitive or disease progression groupings based on H-Y staging or MMSE scores. Despite this, we identified an interesting metabolite in the cognitive and disease progression experiments: carnitine. Carnitine is a long-chain amino acid synthesized in the brain from essential amino acids lysine and methionine, which undergoes acetylation in the mitochondria to form acetyl-L-carnitine (Sergi et al., 2018). Many studies have focused on this compound, which has neuroprotective properties (Afshin-Majd et al., 2017; Gill et al., 2018). Some studies have also found that carnitine can prevent neuronal loss in Parkinson’s rat models and improve memory function (Singh et al., 2018). Recent studies have suggested a specific deficiency of free carnitine in early-stage Alzheimer’s disease (AD) in women. Patients with lower levels of free carnitine exhibit higher accumulation of β-amyloid (Aβ) and increased t-Tau levels in cerebrospinal fluid (CSF) (Bigio et al., 2025). In our results, carnitine levels were observed to be lower in both the severe H-Y stage group and the MCI group (based on MMSE scores), which aligns with the previously noted patterns. The mechanism of carnitine’s function in the body is shown in Figure 10 (Longo et al., 2016; McCann et al., 2021; Virmani and Cirulli, 2022). However, the relatively small sample size in the PD group limits the scope of these findings and warrants further investigation.

In addition, for the metabolomics data of the PD and CON groups, we also used the Random Forest method to build a robust and reliable model, which supported the predictive power of our biomarkers. This model suggested strong predictive performance under double validation, supporting the ability of our biomarkers to predict the disease and indicating a potential for the effectiveness of the biomarkers we identified.

The approach of combining microbiomics with untargeted metabolomics, using TC as a sample to identify biomarkers for neurodegenerative diseases, followed by validation through machine learning models, offers a promising new methodology. Compared to traditional sampling methods such as CSF, blood, urine, or feces, collecting TC is more convenient, less invasive, and more acceptable to patients. Moreover, it has greater potential for widespread adoption in community healthcare settings. If the sampling and processing procedures are thoroughly developed and standardized through research, patients could eventually collect samples themselves for basic testing. These findings suggest a potential foundation for future studies investigating the diagnostic and therapeutic role of TC in PD and other neurodegenerative diseases.

In conclusion, our study underscores the potential of TC samples as a non-invasive, cost-effective tool for diagnosing and monitoring PD. Metabolomics and microbiome analysis revealed differences between PD patients and controls, hinting at potential metabolic and microbial alterations relevant to PD. Key findings include reduced levels of palmitoylethanolamide (PEA) in PD patients, as well as a correlation between specific microbial phyla and differential metabolites. PEA deficiency may contribute to PD pathogenesis, but further studies are needed to fully elucidate its role. Moreover, the decreased carnitine levels observed in the severe H-Y stage and mild cognitive impairment (MCI) groups suggest a possible link to PD progression and cognitive decline. The application of machine learning techniques further supported the robustness of these findings, yielding an accuracy of 88.9%. This suggests that TC samples may hold potential as a tool for early PD diagnosis, a possibility warranting further investigation. These results suggest the clinical potential of combining metabolomics, microbiome analysis, and machine learning in advancing PD diagnostics and monitoring, which could lead to non-invasive, reliable tools for PD management. Most importantly, it offers new approaches and perspectives that could contribute to advancing PD research.

In summary, our findings suggest that conducting metabolomics and microbiomics studies using TC samples from PD patients provides an efficient means to uncover disease characteristics. However, the relatively small sample size and single-source outpatient data may limit the generalizability of our findings. A key aspect of our methodology was the implementation of strict inclusion criteria in our outpatient sampling. Our target population included PD patients committed to regular four-week follow-ups, and post-sample collection, any samples failing cell count standards were excluded. Consequently, only patients who successfully navigated all these rigorous screening and exclusion steps were included in the final study. We hope that through these stringent inclusion and exclusion criteria, our exploratory study, despite its small sample size, can provide a more robust data foundation. Our study supports the potential of TC samples as a non-invasive tool for PD diagnosis, but further validation in larger cohorts and multiple centers is required to confirm its clinical utility. Additionally, the current limitations in metabolite databases may have led to incomplete metabolite identification, highlighting the need for more comprehensive and updated reference databases to improve future research outcomes. The complexity of microbiome data, coupled with the limitations in functional inference, may result in the underrepresentation of certain microbial components. Future improvements could be achieved through more precise functional prediction models, enhancing the accuracy of microbiome analysis. Furthermore, variations in TC sampling methods across different research teams highlight the need for standardized protocols to ensure scientific rigor and consistency. Currently, we are actively strengthening our collaborations with neurologists and specialists. We plan to expand the scope of our research by conducting studies in multiple hospital environments. This strategy is expected to significantly increase our sample size and enhance the generalizability of our research findings. Future studies should focus on further validating the use of TC as a non-invasive sampling material, addressing standardization issues, and expanding its applicability in community healthcare settings.