Amphetamine-type stimulants (ATSs) are a highly prevalent group of illicit drugs with an estimated 27 million global users (Unooda, 2022). Methamphetamine is the most commonly used ATS (Unooda, 2022), and methamphetamine use disorder (MUD) has emerged as a considerable global public health concern (Kohno et al., 2020). MUD is a chronic, relapsing brain disease (Leshner, 1997) that influences the physiology of various systems in patients, such as the central neural system and the gastrointestinal system (Prakash et al., 2017). However, current diagnostic criteria for MUD are based on subjective and qualitative symptoms and feelings reported by patients (Fuchs, 2010; Segata et al., 2011). Although urine and blood tests are commonly used to detect the use of methamphetamine (Cruickshank and Dyer, 2009), the results do not necessarily provide a clear indication of MUD (Segata et al., 2011; Prakash et al., 2017). It is possible for individuals who casually use methamphetamine to test positive, despite not exhibiting MUD. Moreover, some MUD patients may not test positive if they have not used the drug during the test window (Oyler et al., 2002). As such, it is critical to explore alternative objective biomarkers and diagnostic models to more accurately assess the presence of MUD, rather than relying solely on measures of recent drug exposure (Segata et al., 2011).

The withdrawal period plays a pivotal role in MUD, as it is critical for determining a patient’s prognosis regarding whether they will achieve sustained recovery or experience a relapse (Zorick et al., 2010). Specifically, the initial three months after withdrawal are of utmost importance, as MUD patients tend to experience worse outcomes, including elevated levels of anxiety, depression, impulsivity, and cravings for methamphetamine, during this period(Wang et al., 2013). In addition, in MUD patients who were undergoing a withdrawal period of within three months, relative glucose metabolism was observed to be higher in the parietal cortex and lower in the striatum and thalamus (Nora D. Volkow et al., 2001). Furthermore, the first 3 months after withdrawal is a crucial period for intervention (McKetin et al., 2012). Moreover, the duration of withdrawal reflects the time of the most recent methamphetamine use by MUD patients, which may not only provide clues for clinical intervention but also for judicial purposes. However, current detection methods are still inadequate for determining the duration of withdrawal and the time of the most recent drug use by MUD patients.

As research on the microbiota-gut-brain axis continues to expand, the role of the gut microbiota in psychiatric diseases has become increasingly evident (Cryan et al., 2019; Qin et al., 2021). Indeed, the gut microbiota has emerged as a promising new biomarker for a range of central nervous system diseases and mental illnesses, including Parkinson’s disease (Nair et al., 2018), schizophrenia (Zhu et al., 2020), and depression (Hu et al., 2019). Recent research has indicated that compared to healthy controls (HCs), MUD patients exhibit significant differences in some gut microbes (Deng et al., 2021; Yang et al., 2021). The gut microbiota may participate in the development of MUD through its release of inflammatory mediators, such as bacterial lipopolysaccharide (Deng et al., 2021; Yu et al., 2023). Moreover, a previous study conducted on rats showed that gut dysbiosis after methamphetamine cessation changed in relation to the duration of withdrawal (Forouzan et al., 2021). Based on the above evidence, we hypothesized that MUD patients exhibit gut dysbiosis, and this dysbiosis may vary depending on the duration of withdrawal. In addition, these gut microbiota dysbiosis features can help identify MUD patients and the amount of time they have been in withdrawal.

The gut microbiota is a multifaceted and ever-changing community of diverse microbes (Cryan et al., 2019). Given its intricacy, machine learning, a valuable technique for analyzing heterogeneous biological data with inherent noise, has been commonly employed in the development of microbiota-based diagnostic models (Namkung, 2020). In this research, our objective was to examine gut dysbiosis among MUD patients and generate a machine-learning model utilizing gut microbiota features to facilitate the identification of MUD patients. Furthermore, in patients with MUD, we attempted to explore the variability in the gut microbiota in MUD patients who underwent varying durations of abstinence and assess the potential of the microbiota as an indicator for identifying the amount of time spent in withdrawal by MUD patients.

Since previous studies have mainly focused on male MUD patients and involved small sample sizes (Deng et al., 2021; Yang et al., 2021), we conducted the largest data-available study to date to compare the gut microbiota composition in 78 MUD male and female patients and 50 sex- and age-matched HCs. We identified 90 taxa from five phyla as potential biomarkers to help identify MUD patients. Based on these taxa, we developed a machine learning model that was highly effective in distinguishing MUD patients. The model also showed generalizability across regions. To further explore the potential role of the gut microbiota in distinguishing MUD patients with different withdrawal stages, we identified 18 taxa as potential biomarkers, and the model based on the taxa also performed well. To our knowledge, this study is the first to provide evidence for the role of the microbiota in MUD identification and to illustrate the differences in the gut microbiota between MUD patients with varying withdrawal times.

In this study, the gut microbiota of MUD patients showed a significant decrease in evenness and a downward trend in richness compared with that of HCs. Additionally, there was a significant difference in beta diversity between the two groups. These findings indicate that individuals with MUD may have an imbalanced gut microbiota. This is consistent with previous studies (Yang et al., 2021; Wang et al., 2023). In addition to diversity, enterotype is an important indicator of the human gut microbiota (Arumugam et al., 2011). This study is the first to compare the enterotype distribution between MUD patients and HCs, and the results showed that MUD patients have a higher percentage of individuals with the ET_F enterotype. Another study found that MUD patients also have a higher percentage of people with the ET_F enterotype than casual methamphetamine users (He et al., 2023). This suggests that the prevalence of the ET_F enterotype may be associated with MUD. One possible explanation is that the gut environment of MUD patients might be more conducive for bacteria associated with the ET_F enterotype to thrive. However, considering the two-way communication between the gut and the brain, there is also a possibility that individuals with the ET_F enterotype have a higher susceptibility to MUD. The taxonomic distribution results revealed additional details regarding the taxa associated with the ET_F enterotype. Among the top 10 genera, Lachnoclostridium and Lachnospira were significantly enriched in MUD patients, and both were associated with the ET_F enterotype. Lachnoclostridium and Lachnospira are from the family Lachnospiraceae, and their abundance has been found to be positively related to psychotic symptoms in MUD patients, such as difficulty in abstract thinking (Yang et al., 2021). Additionally, it was observed that MUD patients had a significantly higher abundance of Firmicutes and a higher Firmicutes/Bacteroidetes ratio than HCs, which are also features of the ET_F enterotype (Arumugam et al., 2011). An unbalanced Firmicutes/Bacteroidetes ratio is indicative of gut dysbiosis and is associated with various chronic diseases, such as obesity (Magne et al., 2020), inflammation (Stojanov et al., 2020), and hypertension (Yang et al., 2015). Interestingly, MUD patients are also a high-risk population for these diseases (Kaye et al., 2007; J Edge and Gold, 2011; Davidson et al., 2022). Even though it is challenging to determine a causal relationship, future studies should consider the potential role of the gut microbiota in the increased susceptibility to chronic illnesses among MUD patients. In addition, some taxa, including Bacteroides and Prevotella, did not differ between the HCs and MUD patients. These two taxa were interpreted as biomarkers of diet and lifestyle (Gorvitovskaia et al., 2016). This suggests that diet and lifestyle factors may have been similar between the HCs and MUD patients in this study. Although our findings demonstrate a significant association between certain gut bacteria and MUD, we have to acknowledge that the study’s design does not allow us to establish causal relationships between changes in the gut microbiota and the development of MUD. To more effectively explore causality, future longitudinal studies would enable us to track changes in both the gut microbiota and MUD development over time, providing more insight into potential causal relationships. Additionally, future interventional trials could involve manipulating the gut microbiota composition and assessing subsequent changes in MUD incidence, offering a more direct way to investigate causality.

In summary, MUD patients showed a notable imbalance in the gut microbiota. To develop a microbiota-based identification model for MUD, we utilized LEfSe to identify potential biomarkers. LEfSe is a commonly utilized method, based on the machine learning model of LDA, to identify biomarkers in the gut microbiome (Segata et al., 2011). Some of the taxa identified by LEfSe have already been shown to potentially contribute to the pathological processes of MUD. For example, Klebsiella has been found to influence the lipopolysaccharide level of the host and further mediate the craving for methamphetamine after withdrawal in rats (Yu et al., 2023). Parasutterella has been found to be related to the impairment of bile acid homeostasis and host weight gain after methamphetamine use in mice (Zhang et al., 2022). The results related to the gut microbiota features of MUD might provide new perspectives for developing future personalized treatment approaches to reduce drug craving or metabolism imbalance. Based on these potential biomarkers, we developed a machine learning model that performed well (accuracy = 0.894) to help identify MUD patients. There has been no previous study on a microbiota-based model to help identify MUD. What we known about ability of microbiota as the potential biomarker upon previous study is largely based on other diseases. Based on gut microbial biomarkers, Addolorato et al. developed a random forest classifier with an accuracy of 0.934 for identifying patients with alcohol use disorder compared to HCs (Addolorato et al., 2020). Based on oral microbial biomarkers, Kosciolek et al. developed a machine learning model with an accuracy of 0.83 for identifying patients with substance use disorder (including both stimulants and opioids) compared to HCs (Kosciolek et al., 2021). The high-performance metrics of the models were consistent with previous studies on microbial biomarkers for addiction diseases. Considering addiction is a chronic brain disease, another popular potential biomarker is neuroimaging (Yang et al., 2023). In contrast to previous neuroimaging-based models for identifying MUD patients, which had an accuracy range of 0.732-0.880 (Yu et al., 2020; Yan et al., 2021), microbial-feature-based models demonstrate more satisfactory efficacy. Additionally, the generalizability across regions of the model adds another layer of significance. To validate the efficacy and generalizability of our MUD identification model, we introduced an external dataset. Nonetheless, it is crucial to recognize that discrepancies in population demographics, sample collection methods, or laboratory protocols across datasets can collectively influence the model’s performance when extrapolated to an external dataset. For instance, in the external dataset, all samples consisted of male individuals, potentially accounting for the observed decrease in accuracy when the model was applied to this external dataset. However, although the model’s performance when applied to an external dataset declined, it continued to demonstrate promising classification capabilities. The shared effective microbial biomarkers between cohorts from Northwest China (Shenyang) and Central China (Wuhan) in this study suggest that while distinct regional populations might exhibit varying microbial community structures, potential biomarkers for microbial dysbiosis associated with MUD could possess universal attributes. These potential biomarkers manifest a consistent trend in distribution and characteristics of the gut microbiota in MUD patients (Yang et al., 2021; Wang et al., 2023). The results indicate that the microbial changes in MUD patients might be stable across different regions, and the combination of microbiota features and the machine learning model could be an innovative strategy for exploring noninvasive biomarkers for MUD. This approach might help address the issue of subjectivity in current diagnosis techniques based on symptoms (Fuchs, 2010) as well as the limited time window for urine tests (Cruickshank and Dyer, 2009).

While the model’s metrics showcased satisfactory performance, it is necessary to deliberate on the practical significance of its predictive capacity with caution. The inherent limitations of employing machine learning models for clinical outcome prediction cannot be overlooked. First, machine learning models are inherently constrained by the training dataset. Considering that the gut microbiota is notably impacted by ethnicity and environment, even if the model demonstrates strong performance on an external dataset, the applicability of the model to broader contexts is subject to scrutiny. Moreover, machine learning models are frequently regarded as ‘black boxes’ due to the lack of transparency and interpretability. In the realm of clinical practice, it is imperative for the benefit of patients that psychiatrists grasp the underlying rationale for every clinical decision. The models should be regarded as complementary tools, designed to supplement rather than supplant the judgment and expertise of psychiatrists.

To further analyze whether the gut microbiota of MUD patients with different withdrawal times exhibit distinct features, we compared the gut microbiota between long-/short-term-withdrawal MUD patients. Our findings revealed a significant difference in beta diversity and the levels of some taxa, such as Collinsella, between long- and short-term withdrawal MUD patients, suggesting that the gut microbiota of MUD patients may change during withdrawal. A previous study conducted on rats also showed that methamphetamine cessation significantly changed the composition of the gut microbiota, and these changes fluctuated depending on the amount of time after cessation (Forouzan et al., 2021). A human research study (Deng et al., 2021) including MUD patients with a longer withdrawal time (an average of 8 months) reported different results in both gut microbiota diversity and composition from studies that included MUD patients with a withdrawal time ranging from 1 week to 6 months (Cook et al., 2019; Yang et al., 2021). Discerning differences among MUD patients who have undergone different withdrawal periods may prove instrumental in identifying the time of the most recent methamphetamine use and recent methamphetamine usage. This is crucial for accurately assessing and further helping provide better MUD management strategies. Our model confirmed that the microbiota differed in MUD patients with different withdrawal periods and indicated that microbiota features may be potential biomarkers to help identify the time of the most recent methamphetamine use by MUD patients, which may further help prompt drug use information in the clinic. However, considering that the sample size of the training dataset was not satisfactory, the model had a risk of overfitting. The result of the MUD withdrawal period identification model should be interpreted with caution.

We acknowledge that there were several limitations to this study. First, the differences in the gut microbiota between long- and short-term-withdrawal MUD patients were based on cross-sectional data, and host factors might have confounded the results. Second, the results of the methamphetamine withdrawal period identification model are exploratory due to the limited sample size for training and lack of an external test dataset. Third, we need to consider the limitations of the study design and specific study populations. Although the results regarding Bacteroides and Prevotella suggested that diet and lifestyle factors may be similar between HCs and MUD patients, specific information on diet and lifestyle was not collected in this study design. Moreover, the samples included in this study were limited to MUD patients. The results cannot be generalized to other substance use disorders. Fourth, although we analyzed the enterotypes of samples with the enterotype classifier based on the reference dataset of the European Metagenomics of the Human Intestinal Tract project to ensure consistency with previous studies of enterotypes based on large samples, this classification system for gut microbiome enterotypes is still debatable. Additional study on the enterotypes of MUD is needed with the further development of human enterotype research.

In summary, the gut microbiota of MUD patients and HCs differ significantly, and the gut microbiota of MUD patients may change during methamphetamine withdrawal. We developed machine learning models that effectively distinguish MUD patients and determine their withdrawal period, which demonstrated the potential of using gut microbiota as biomarkers in identifying MUD patients. Overall, our results suggest that the fecal microbiome-based model for identifying MUD patients is feasible, which has important implications for the development of noninvasive diagnostics and the establishment of a comprehensive MUD disease management strategy. Future longitudinal studies with larger sample sizes are needed to validate these findings, and multiomics solutions are needed to explore the underlying mechanisms.

The data presented in the study are deposited in the Sequence Read Archive, accession number PRJNA970410 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA970410).

The studies involving humans were approved by the Ethics Committee of the First Hospital of China Medical University. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

LL: Conceptualization, Data curation, Formal Analysis, Investigation, Methodology, Project administration, Visualization, Writing – original draft. ZD: Conceptualization, Data curation, Investigation, Project administration, Writing – review & editing. WL: Investigation, Project administration, Writing – review & editing. RL: Methodology, Supervision, Writing – review & editing. TM: Investigation, Writing – review & editing. YZ: Conceptualization, Supervision, Writing – review & editing. EW: Investigation, Writing – review & editing. YT: Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing – review & editing.