Abstract

Background

   There is growing research on the relationship between gut bacteria and
   various forms of strokes. This study aimed to investigate the
   relationship between fecal metabolites and ischemic stroke, providing a
   new perspective on predicting the latter.

Results

   Stool samples were taken from 60 patients with ischemic stroke and 60
   healthy individuals, and non-targeted metabolomic analysis was used.
   The generalized boosted linear model was utilized for co-occurrence
   analysis to ascertain the noteworthy variation in fecal metabolites.
   The important differential metabolites were identified by the random
   forest algorithm, a prediction panel was developed to distinguish
   ischemic stroke patients from healthy individuals. Specifically, six
   differential metabolites (Ganoderic acid theta, Fructose-lysine,
   Pentaethylene glycol, 2-Chlorooctadecanoic acid, PA(2:0/PGF1alpha), and
   4-[(E)-5,6-Dihydro-2,3'-bipyridin-3(4H)-ylidenemethyl]-3-methoxyphenol)
   were identified as potential independent stroke-associated metabolites.
   A prediction panel consisting of these six metabolites could yield an
   area under the curve of 0.989 in training set and 0.973 in testing set.
   There was a substantial correlation between all six independent
   stroke-associated metabolites and the severity of ischemic stroke, but
   it was not affected by depression or anxiety.

Conclusions

   These six differential metabolites were independent stroke-associated
   metabolites, and the panel consisting of these metabolites could serve
   as a potential prediction panel for ischemic stroke. However, future
   external validation in multi-ethnic cohorts is necessary to confirm
   broader generalizability.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s12866-025-04217-8.

   Keywords: Metabolome, Ischemic stroke, Gut, Biomarker

Background

   A complex series of neurological, pathological, and physiological
   processes culminate in brain damage caused by ischemic stroke [[46]1,
   [47]2], and its frequency is continuously rising globally [[48]3].
   Effective drugs or therapies are currently unavailable for individuals
   who missed the thrombolysis and surgical embolization therapy windows
   [[49]4]. Thus, there is an urgent need for efficient alternative
   therapeutic approaches to encourage brain function restoration and
   lessen the severity of impairment linked to ischemic stroke [[50]5].

   The gut microbiota preserves physical health and is intrinsically
   connected to the emergence of several illnesses [[51]6–[52]8]. The
   enteric and central nervous systems can communicate in both directions;
   this is known as the gut-brain axis of the gut microbiota
   [[53]9–[54]11]. More than half of ischemic stroke patients experience
   gastrointestinal problems, and a disturbed gut microbiota may increase
   the risk of stroke, impact post-stroke mortality, and cause progressive
   neurological impairments [[55]12, [56]13]. Recent evidence has also
   suggested a key role of the gut microbiota and its metabolites in
   ischemic stroke [[57]14, [58]15]. Enrichment with Enterobacteriaceae
   has the potential to increase inflammation and worsen brain infarction,
   which is a risk factor for poor outcomes for stroke patients on its own
   [[59]16]. However, it is still mostly unclear how intestinal flora
   contributes to ischemic stroke.

   Following an ischemic stroke, disease progression picks up rapidly
   [[60]17, [61]18]. The metabolites generated in the plasma during the
   beginning of ischemic stroke should be examined in addition to
   neurological abnormalities and imaging results to guarantee prompt
   identification and clarify new biological mechanisms linked to
   functional recovery [[62]19]. The intricacy of the ischemic cascade,
   the effect of the blood–brain barrier on the diffusion of blood
   biomarkers, and differing interpretations of laboratory results have
   all contributed to the restricted use of blood biomarkers of stroke.
   S-100 protein, brain natriuretic peptides, fibrinogen, and serum-free
   hemoglobin have all been demonstrated to be potential stroke biomarkers
   in recent research; however, none of these biomarkers have demonstrated
   enough sensitivity or specificity for clinical usage [[63]20].

   Non-targeted metabolomics is a promising method for investigating
   biomarkers and their mechanisms in cerebrovascular diseases [[64]21].
   Brain illness is increasingly being studied using patient peripheral
   tissues, such as fecal matter, and some metabolites found in these
   studies may be useful as potential disease biomarkers [[65]22, [66]23].
   Thus, fecal research is necessary to identify the important metabolites
   in ischemic stroke and determine whether they have any potential as
   biomarkers for the objective diagnosis of ischemic stroke. In this
   study, we assessed the metabolomic analysis of feces in individuals
   with ischemic stroke patients. Finding possible biomarkers and
   metabolic changes in ischemic stroke patients was the main aim of our
   present investigation. Additionally, fecal metabolites that might be
   useful therapeutic targets and have implications for stroke therapy are
   shown by our investigation.

Methods

Human subjects

   This study was a hospital-based case–control study. The research has
   been carried out in accordance with the World Medical Association
   Declaration of Helsinki. Ethical approval was obtained from the
   Institutional Ethics Committee of the First Affiliated Hospital of
   Chongqing Medical University (2023–414). Written informed consent was
   secured from all participants prior to stool sample collection. Between
   February 2023 and September 2023, inpatients with ischemic stroke were
   enrolled from the Department of Neurology, the First Affiliated
   Hospital of Chongqing Medical University, while HCs were recruited from
   the hospital’s Medical Examination Center. All participants were Han
   Chinese residents of Chongqing, China. Inclusion criteria for ischemic
   stroke patients included: (a) aged 18 to 70 years; (b) acute
   hospitalization within 48 h after onset; (c) confirmed cerebral
   infarction via intracranial CT or MRI. Exclusion criteria for ischemic
   stroke patients were as follows: (a) history of transient ischemic
   attacks, intracerebral hemorrhage, transient cerebral ischemia, or
   other brain diseases; (b) severe dysfunction of heart, liver, kidney,
   digestive system, or other organs; (c) psychiatric disorders, cancer,
   or dementia; (d) pregnancy, breastfeeding, or planned pregnancy; (e)
   use of anticoagulants, lipid-lowering drugs, probiotics, or antibiotics
   within the past month; (f) recent infection within the past week. For
   HCs, exclusion criteria included: (a) history of stroke or neurological
   disorders; (b) diagnosis of cancer, metabolic, cardiovascular,
   psychiatric, gastrointestinal, cerebrovascular diseases, or dementia;
   (c) age < 18 or > 70 years; (d) pregnancy, breastfeeding, or planned
   pregnancy; (e) intake of anticoagulants, lipid-lowering drugs,
   probiotics, or antibiotics within the past month; (f) recent infection
   within the past week.

Clinical assessments and the gathering of samples from human subjects

   Clinical baseline data were collected, including age, gender, and BMI.
   The NIHSS score evaluates stroke severity based on clinical
   characteristics [[67]24]. Given the well-documented association between
   ischemic stroke and increased risk of depression and anxiety, which can
   significantly impair daily functioning and quality of life [[68]25], we
   also assessed the impact of these psychological symptoms. Severity of
   depressive and anxious symptoms was measured using the HAMD
   [[69]26–[70]29] and the HAMA [[71]30, [72]31]. A HAMA score > 7
   indicated the presence of anxiety, while a HAMD score > 17 was used to
   define depression.

   Fecal samples were collected immediately upon patient admission to the
   hospital, prior to administering any ischemic stroke medications to
   eliminate potential confounding effects of post-stroke pharmacotherapy
   on the metabolomic profile. All samples were promptly immersed in
   absolute ethanol, transported on ice to the laboratory within 2 h, and
   subsequently stored at −80℃ until analysis.

Untargeted metabolomics analysis

   Fecal samples were processed using homogenization, dissociation, and
   centrifugation, following our previously established protocols
   [[73]32–[74]34]. Briefly, fecal homogenization was performed as
   follows: approximately 50 mg of feces were weighed, mixed with 400μL
   ice-cold extraction solvent (80% methanol), and homogenized using a
   bead-beater. The supernatant was collected after centrifugation, and
   the metabolic profile was analyzed by LC–MS/MS. An internal standard
   (0.02 mg/mL L-2-chlorophenylalanine) was spiked into fecal extracts
   before LC–MS/MS analysis. Metabolite intensities were normalized to the
   internal standard to correct for extraction efficiency. The
   normalization efficiency was also validated by principal component
   analysis. Moreover, to ensure system stability and data reliability, a
   pooled quality control (QC) sample was prepared by combining equal
   volumes (20 μL) of each sample supernatant. QC samples were processed
   and analyzed identically to experimental samples throughout the
   workflow.

   Using an ACQUITY HSS T3 column (100 mm × 2.1 mm i.d., 1.8 μm; Waters,
   USA), a Thermo UHPLC-Q Exactive HF-X system was utilized for the
   untargeted liquid chromatography-mass spectrometry/mass spectrometry
   (LC–MS/MS) metabolomics investigation at Majorbio Bio-Pharm Technology
   Co. Ltd. (Shanghai, China). For positive electrospray ionization
   (ESI +) mode, mobile phases consisted of 0.1% formic acid in water
   (solvent A) and 0.1% formic acid in acetonitrile (solvent B). In
   negative ESI-mode, solvents were 6.5 mM ammonium bicarbonate in water
   (solvent C) and 6.5 mM ammonium bicarbonate in 95% methanol (solvent
   D). Chromatographic conditions were set as follows: flow rate
   0.40 mL/min, column temperature 40 °C, and injection volume 2 μL. The
   mass spectrometer operated in dual-mode ESI with optimized parameters:
   source temperature 425 °C, sheath gas flow 50 arb, auxiliary gas flow
   13 arb, spray voltage − 3500 V (ESI −)/+ 3500 V (ESI +), and normalized
   collision energy 20–40-60 V for MS/MS fragmentation.

   Using Progenesis QI (Waters Corporation), baseline filtering, peak
   identification, integration, retention time correction, peak alignment,
   and normalization were applied to the raw data produced by LC–MS/MS
   [[75]35]. The Human Metabolome Database ([76]http://www.hmdb.ca),
   Metlin ([77]https://metlin.scripps.edu), and self-built databases were
   used to identify metabolites. The highest intensity was determined to
   be the metabolite expression level [[78]36]. The difference in fecal
   metabolites between the two groups was determined using OPLS-DA. To
   assess the developed OPLS-DA model’s goodness-of-fit and
   predictability, two parameters (R^2X and R^2Y) were used. In the
   meantime, a 300-iteration permutation test was run to determine whether
   the distance between ischemic stroke patients and HCs was not randomly
   distributed. The resulting OPLS-DA loading plot indicated that the
   differential fecal metabolites responsible for discriminating between
   the two groups were those with VIP > 1.0 and p < 0.05. Then, the
   detected differential fecal metabolites were used to perform pathway
   enrichment analysis based on the Kyoto Encyclopedia of Genes and
   Genomes (KEGG) database ([79]http://www.genome.jp/kegg/).

Statistical analysis

   The R 4.0.5 software was utilized for all the analyses, and where
   necessary, the Student’s t-test, Chi-squared test, and Spearman
   correlation analysis were applied. The method of Benjamini and Hochberg
   False Discovery Rate was used to conduct multiple testing corrections,
   and the threshold of adjusted p-value was 0.05. Co-occurrence analysis
   was performed using the GBLM to determine the significantly different
   fecal metabolites. In short, within the constructed network,
   metabolites were categorized as crucial (i.e., stress centrality: the
   importance of metabolites to maintain the network’s integrity) and
   influential (i.e., betweenness centrality: the direct and indirect
   influences of metabolite on other metabolites). After that, a simpler
   prediction panel for dividing ischemic stroke patients from HCs was
   created by further analyzing the significant differential fecal
   metabolites that had been detected using the random forest algorithm.
   The capacity of this panel to differentiate between ischemic stroke
   patients and HCs was measured using the ROC curve analysis [[80]37,
   [81]38].

Results

Fecal metabolites alterations in ischemic stroke patients

   A total of 416 subjects were screened, and 120 subjects (60 ischemic
   stroke patients and 60 healthy controls (HCs)) were included in the
   final dataset (Supplementary Fig. S1). The clinical and demographic
   details of the individuals that were recruited are compiled in
   Supplementary Table S1. The results of Orthogonal Partial Least-Squares
   Discriminant Analysis (OPLS-DA) model showed that there was no overlap
   between the two groups and the values of R^2X (0.95) and Q^2 (0.71)
   permutation test (n = 399) quantifying the built model was positive
   (Fig. [82]1A), suggesting a robust metabolic difference between
   ischemic stroke patients and HCs. Moreover, the results of the
   permutation test showed that the blue regression line of Q2-points
   intersected the vertical axis (on the left) below zero (Q2 = (0,
   −0.58)), further demonstrating the validity of the built OPLS-DA model
   (Fig. [83]1B).

Fig. 1.

   [84]Fig. 1
   [85]Open in a new tab

   Metabolomic analysis of fecal samples from ischemic stroke patients and
   HCs. A OPLS-DA model built using fecal metabolites showed that ischemic
   stroke patients were separately from HCs with no overlap, suggesting
   the divergent fecal metabolic phenotypes between the two groups; B the
   results of 300-iteration permutation test showed that the built model
   was robust (Q2 = (0, −0.58)); C 202 differential metabolites
   (VIP > 1.0, p < 0.05, FDR < 0.05) were identified according to the
   corresponding OPLS-DA loading plot; D classification of differential
   metabolites showed that the lipids and lipid-like molecules category
   ranked at the top, with a proportion of 31.19%. HCs, healthy controls;
   FDR, false discovery rate; OPLS-DA, orthogonal partial least-squares
   discriminant analysis; VIP, variable importance in projection

   By analyzing the corresponding OPLS-DA loading plot, 96 up-regulated
   and 106 down-regulated differential fecal metabolites (variable
   importance in projection (VIP) > 1.0, p < 0.05, false discovery rate
   (FDR) < 0.05) were identified (Fig. [86]1C). Among these differential
   metabolites, 63 (31.19%) metabolites belonged to lipids and lipid-like
   molecules (Fig. [87]1D). The detailed information of these differential
   metabolites was described in Supplementary Table S2.

Enrichment analysis of differential fecal metabolites in ischemic stroke
patients

   As shown in Fig. [88]2A, the hierarchical clustering heat-map
   constructed using these identified differential fecal metabolites
   showed a consistent clustering pattern within the individual groups,
   which suggested that the most significant deviations between ischemic
   stroke patients and HCs could be described by these differential fecal
   metabolites.

Fig. 2.

   [89]Fig. 2
   [90]Open in a new tab

   Pathway enrichment analysis of differential fecal metabolites. A A
   heat-map analysis of differential fecal metabolites between the two
   groups was shown; B Pathway enrichment analysis showed that 20 pathways
   had p < 0.05, and only eight of them had FDR < 0.05: Tyrosine
   metabolism, Nucleotide metabolism, Synaptic vesicle cycle, ABC
   transporters, Neuroactive ligand-receptor interaction, Bacterial
   chemotaxis, Mineral absorption, and Taste transduction. FDR, false
   discovery rate

   Pathway enrichment analysis was conducted here to explore the potential
   functions of these differential metabolites. The results showed that
   these differential metabolites were significantly enriched in 20
   pathways (p < 0.05), but only eight of them had FDR < 0.05: Tyrosine
   metabolism (FDR = 0.0001), Nucleotide metabolism (FDR = 0.0027),
   Synaptic vesicle cycle (FDR = 0.0025), ABC transporters (FDR = 0.0105),
   Neuroactive ligand-receptor interaction (FDR = 0.0099), Bacterial
   chemotaxis (FDR = 0.0125), Mineral absorption (FDR = 0.0160) and Taste
   transduction (FDR = 0.0188) (Fig. [91]2B). There are seven differential
   metabolites in Tyrosine metabolism, which was the highest among all the
   pathways. The detailed information on these eight significantly
   enriched pathways was described in Supplementary Table S3.

Important differential fecal metabolites in ischemic stroke patients

   Spearman correlation analysis was used to explore the potential
   relationships between differential fecal metabolites and clinical
   indexes, including age, body mass index (BMI), Hamilton Anxiety Rating
   Scale (HAMA), 17-item Hamilton Rating Scale of Depression (HAMD), and
   National Institutes of Health Stroke Scale (NIHSS). The results showed
   that the differential fecal metabolites were mainly correlated with
   NIHSS; and they also had some impacts on HAMA and HAMD, but had little
   relationship with age and BMI (Fig. [92]3A).

Fig. 3.

   [93]Fig. 3
   [94]Open in a new tab

   Important differential fecal metabolites in ischemic stroke patients. A
   The results of correlation analysis showed that differential
   metabolites were mainly correlated with NIHSS, and have some and little
   relationships with HAMA/HAMD and age/BMI, respectively. B GBLM showed
   that 12 differential metabolites that had higher influence and critical
   places in the network were viewed as important differential fecal
   metabolites in stroke patients. Red and blue indicated positive and
   negative correlations, respectively; and the thicker the line, the
   bigger the correlation coefficient. BMI, body mass index; GBLM,
   Generalized boosted linear model; HAMA, Hamilton Anxiety Scale; HAMD,
   Hamilton Depression Scale; NIHSS, National Institutes of Health Stroke
   Scale

   Among these differential fecal metabolites, 12 metabolites were
   identified as important differential fecal metabolites by Generalized
   Boosted Linear Model (GBLM) analysis (Fig. [95]3B):
   4-[(E)−5,6-Dihydro-2,3'-bipyridin-3(4H)-ylidenemethyl]−3-methoxyphenol,
   Pentaethylene glycol, Fructose-lysine, Calystegine A6,
   Trans-4-Aminocyclohexanecarboxylic acid, Dihydrocitrinone,
   Cyclopentanone, Muricatin C, Estradiol benzoate, Ganoderic acid theta
   (GAT), PA(2:0/PGF1alpha), and 2-Chlorooctadecanoic acid (CA). Moreover,
   the last five differential fecal metabolites belonged to lipids and
   lipid-like molecules, and four metabolites (Pentaethylene glycol (PG),
   Fructose-lysine (FL),
   4-[(E)−5,6-Dihydro-2,3'-bipyridin-3(4H)-ylidenemethyl]−3-methoxyphenol
   (ED), and Trans-4-Aminocyclohexanecarboxylic acid were significantly
   correlated with lipid metabolism.

Independent stroke-associated metabolites

   To find out the independent stroke-associated metabolites, these 12
   identified important differential fecal metabolites were further
   analyzed. Firstly, the included samples were randomly assigned to a
   training set (40 ischemic stroke patients and 40 HCs) and a testing set
   (20 ischemic stroke patients and 20 HCs). The training set was used to
   build a prediction panel, and the testing set was used to independently
   validate the built prediction panel. Secondly, in training set, after
   adjusting age, sex, and BMI, the results of the random forest algorithm
   showed that the following six differential metabolites could be the
   independent stroke-associated metabolites: GAT, FL, PG, CA,
   PA(2:0/PGF1alpha) and ED. The mass to charge ratio (m/z), retention
   time, fragmentation score, formula, and MS/MS spectra of these six
   metabolites are shown in Supplementary Table S4 and Figure S2. Using R
   4.0.5 software, we conducted power calculation (post hoc) using these
   six independent risk factors, and the results showed that our included
   sample size could obtain power = 0.87 (alpha = 0.05). Relative levels
   of these metabolites in two groups are shown in Fig. [96]4A-F. The
   metabolite levels, odd rations, and 95% confidence interval of these
   six stroke-associated metabolites are shown in Supplementary Table S5.

Fig. 4.

   [97]Fig. 4
   [98]Open in a new tab

   Independent stroke-associated metabolites. Six independent
   stroke-associated metabolites were obtained using the random forest
   algorithm. The relative abundance of GAT (A), FL (B), PG (C), ED (D),
   CA (E), and PA(2:0/PGF1alpha) (F) in ischemic stroke patients and HCs.
   The Student’s t-test was used. CA, 2-Chlorooctadecanoic acid; ED,
   4-[(E)−5,6-Dihydro-2,3'-bipyridin-3(4H)-ylidenemethyl]−3-methoxyphenol;
   FL, Fructose-lysine; GAT, Ganoderic acid theta; HCs, Healthy controls;
   PG, Pentaethylene glycol

   In random forest (RF) algorithm analysis, the included samples were
   randomly assigned to a training dataset (40 ischemic stroke patients
   and 40 HCs) and a testing dataset (20 ischemic stroke patients and 20
   HCs). The training dataset was used to train the RF model and the
   testing dataset was used to evaluate the RF model. The built RF model
   could yield an area under the curve (AUC) of 0.998
   (95%CI = 0.993–1.000) in training set and 0.943 (95%CI = 0.873–1.000)
   in testing set. These results showed that the built RF model was
   validated. The prediction panel consisting of these six metabolites was
   as follows:
   P(stroke) = 1/(1 + EXP(−0.065*GAT-4.173*PG-0.463*ED-1.075*FL + 8.998*CA
   -1.307*PA(2:0/PGF1alpha)−16.584)). Receiver operating characteristic
   (ROC) curve analysis showed that this prediction panel could yield an
   AUC of 0.989 (95%CI = 0.970–1.000) in the training set (Fig. [99]5A).
   Moreover, in the testing set, this prediction panel could also yield an
   excellent diagnostic performance for ischemic stroke (AUC = 0.973
   (95%CI = 0.938–1.000); Fig. [100]5B).

Fig. 5.

   [101]Fig. 5
   [102]Open in a new tab

   Diagnostic performances of six fecal metabolites in prediction ischemic
   stroke. Area under the receiver operating characteristic curve (AUC)
   value of the combination of six metabolites (Ganoderic acid theta,
   Fructose-lysine, Pentaethylene glycol,
   4-[(E)−5,6-Dihydro-2,3'-bipyridin-3(4H)-ylidenemethyl]−3-methoxyphenol,
   2-Chlorooctadecanoic acid, and PA(2:0/PGF1alpha)) in diagnosing
   ischemic stroke patients in training (A) and testing sets (B). The
   training and testing sets were randomly selected from independent
   datasets

Assistant analysis of independent stroke-associated metabolites

   These identified independent stroke-associated metabolites had no
   significant correlations with age, sex, BMI, HAMA, and HAMD. However,
   all six independent stroke-associated metabolites were significantly
   correlated with NIHSS: NIHSS was positively correlated with GAT
   (r = 0.41, p = 3.47E-06), FL (r = 0.42, p = 2.52E-07), PG (r = 0.51,
   p = 1.86E-09), ED (r = 0.40, p = 4.85E-06), CA (r = 0.41,
   p = 3.47E-06), and PA(2:0/PGF1alpha) (r = 0.44, p = 5.84E-07), and
   negatively correlated with CA (r = −0.41, p = 3.98E-06) (Fig. [103]6A).

Fig. 6.

   [104]Fig. 6
   [105]Open in a new tab

   Correlations between independent stroke-associated metabolites and
   NIHSS/HAMA/HAMD. A The independent stroke-associated metabolites were
   significantly correlated with NIHSS; B The relative abundances of these
   predictors were similar between ischemic stroke patients with anxiety
   symptoms and ischemic stroke patients without anxiety symptoms; C The
   relative abundances of these predictors were similar between ischemic
   stroke patients with depressive symptoms and ischemic stroke patients
   without depressive symptoms. ED,
   4-[(E)−5,6-Dihydro-2,3'-bipyridin-3(4H)-ylidenemethyl]−3-methoxyphenol;
   HAMA, Hamilton Anxiety Scale; HAMD, Hamilton Depression Scale; NIHSS,
   National Institutes of Health Stroke Scale

   The relative abundances of these six stroke-associated metabolites were
   similar between ischemic stroke patients with anxiety symptoms and
   ischemic stroke patients without anxiety symptoms (Fig. [106]6B).
   Meanwhile, the relative abundances of these stroke-associated
   metabolites were similar between ischemic stroke patients with
   depressive symptoms and ischemic stroke patients without depressive
   symptoms (Fig. [107]6C).

Discussion

   Our study first showed that six differential metabolites (GAT, FL, PG,
   CA, PA(2:0/PGF1alpha), and ED) could be independent stroke-associated
   metabolites. Meanwhile, the panel consisting of these metabolites could
   serve as a potential prediction panel for ischemic stroke. In addition,
   NIHSS was positively correlated with GAT, FL, PG, ED, CA, and
   PA(2:0/PGF1alpha), and negatively correlated with CA. Our results
   suggested that these six metabolites in fecal might play an important
   role in the pathophysiology of ischemic stroke.

   A growing body of research indicates that the incidence of ischemic
   stroke may be influenced by the microbiota and its metabolome [[108]39,
   [109]40]. Thus, it is a potential sector to use the gut microbiota and
   its metabolites as screening techniques for early identification of
   ischemic stroke. Omics approaches hold great potential for the
   development of biomarkers because they can produce large-scale,
   high-throughput data sets with thousands of variables. These data sets
   may be predictive, making them a profitable approach when applied to
   biological systems with extremely high levels of complexity, such as
   the diagnosis of stroke from fecal matter [[110]41, [111]42]. When it
   comes to direct interactions between genetic, environmental, and
   nutritional variables, the fecal metabolome is superior to omics
   techniques that use biofluids like serum and urine [[112]43].
   Therefore, metabolomics studies using fecal samples may find biomarkers
   more successfully. In addition to offering insights into the
   relationships between communities from a microbiome sequencing
   technique, the global nontargeted metabolomics analysis performed here
   on a fecal sample may make it possible to identify metabolic biomarkers
   derived from ischemic stroke.

   Following an untargeted approach, our study identified 96 up-regulated
   and 106 down-regulated differential fecal metabolites in the ischemic
   stroke group, 63 (31.19%) belonged to lipids and lipid-like molecules.
   Previous research has shown that dyslipidemia is widely thought to be a
   key factor in the etiology of ischemic stroke [[113]44, [114]45].
   Baseline levels of triglycerides, very low-density lipoprotein (VLDL)
   triglyceride, VLDL particle number, VLDL size, low-density lipoprotein
   particle number, high-density lipoprotein (HDL) cholesterol, and HDL
   size were also linked to future ischemic stroke in a prospective
   evaluation of lipid and lipoprotein biomarkers associated with incident
   ischemic stroke [[115]46]. In general, our global nontargeted
   metabolomics analysis of fecal samples from ischemic stroke patients
   and HCs revealed distinct and varied metabolic fingerprints. In
   contrast to blood lipid biomarkers, fecal lipid metabolites reflect the
   microbial-mediated transformation of dietary lipids, potentially
   influencing stroke pathogenesis through intestinal barrier dysfunction
   and neuroinflammatory pathways [[116]47]. This mechanistic link to the
   gut-brain axis highlights a novel paradigm, as such microbial-metabolic
   interactions remain underexplored in traditional blood-based biomarker
   research. Our findings demonstrate that fecal lipid dysregulation is
   associated with ischemic stroke, offering new insights into gut-brain
   metabolic crosstalk in cerebrovascular disease.

   Multiple biomarkers on a biomarker panel rather than just one might
   reduce the impact of variance across populations and subgroups, leading
   to more accurate results. Previous researches [[117]48, [118]49]
   indicate that the diagnostic efficacy of combining numerous biomarkers
   was much higher than that of using each biomarker alone. Our earlier
   research [[119]27, [120]50] also showed similar results. Here, we
   discovered that, when used as independent biomarkers for ischemic
   stroke detection, the fecal levels of six metabolites (GAT, FL, PG, CA,
   PA(2:0/PGF1alpha), and ED) demonstrated exceptional diagnostic
   accuracy. Moreover, we found that the discriminative model consisting
   of these six metabolites was able to provide a training set AUC of
   0.989 which was higher. In the meanwhile, the diagnostic robustness of
   this discriminative model was demonstrated by its ability to provide an
   AUC of 0.973 in the testing set. The assessment index used was the AUC;
   values of 1–0.9, 0.9–0.8, 0.8–0.7, 0.7–0.6, and 0.6–0.5 denoted
   classification performance that was excellent, good, fair, poor, or
   fail, respectively [[121]26, [122]28, [123]51]. As a result, the
   discriminative model made up of these six metabolites may be an
   “excellent” classifier of ischemic stroke patients and HCs.

   Following an ischemic stroke, neuropsychiatric problems are common and
   negatively affect functional results and quality of life [[124]52,
   [125]53]. Meanwhile, poststroke depression is thought to be these
   illnesses’ most common clinical manifestation [[126]54]. Because it
   influences outcomes, it is crucial to differentiate between ischemic
   stroke and poststroke depression. Despite this, we discovered that
   these metabolites were only connected with NIHSS in our investigation
   and not with HAMD or HAMA. However, the study’s finding is that
   measurements of these six metabolites have a good discriminant power
   for identifying individuals who are at risk of ischemic stroke. These
   fecal indicators, however, are stable and unaffected by emotional
   swings.

   This study has several inherent limitations that warrant consideration.
   First, the absence of isotope-labeled dietary intervention
   studies—logistically challenging in human populations—hinders
   definitive evaluation of fecal metabolite origins. A substantial
   proportion of gut metabolites arise from dietary intake, gut microbiota
   activity, medications, and lifestyle factors, making it difficult to
   distinguish between host-derived and microbially produced metabolites.
   Second, fecal samples were collected at a single time point, precluding
   the capture of dynamic metabolic changes across the stroke continuum.
   Consequently, observed associations reflect acute-phase alterations
   rather than longitudinal patterns, limiting interpretation to
   early-stage biomarkers rather than stage-independent risk indicators.
   Future longitudinal studies with serial sampling across acute,
   subacute, and chronic phases are needed to characterize temporal
   metabolite profiles and their association with stroke progression,
   recovery, or recurrence. Third, the relatively small sample size
   increases the risk of false-positive findings and limits
   generalizability. While our results are statistically significant,
   validation in larger, multicenter cohorts is imperative to confirm the
   robustness of identified metabolites. Fourth, causal inference remains
   challenging due to the bidirectional relationship between stroke and
   gut physiology: large ischemic lesions can disrupt gastrointestinal
   motility and induce bacterial overgrowth, potentially altering the gut
   metabolome [[127]35]. Thus, whether the observed metabolite changes are
   antecedent stroke-associated metabolites or post-stroke adaptations
   remains unclear. Longitudinal cohort studies with pre-stroke baseline
   sampling are needed to disentangle causality and differentiate between
   preexisting biomarkers and disease-induced adaptations. Fifth, the
   geographic homogeneity of the study population—predominantly Han
   Chinese residents of Chongqing—may limit generalizability to diverse
   ethnic groups or populations with distinct dietary habits, future
   external validation in multi-ethnic cohorts is necessary to confirm
   broader generalizability. Sixth, although correlations between
   differential metabolites and NIHSS scores were observed, these
   associations do not establish direct involvement in ischemic stroke
   pathogenesis. Mechanistic studies are required to elucidate how these
   metabolites are involved in ischemic stroke. Collectively, while our
   findings generate novel hypotheses about gut metabolite-stroke
   associations, they should be interpreted within these methodological
   constraints. Future research addressing sample size, longitudinal
   design, ethnic diversity, and causal pathways will be critical to
   translating these observations into clinical applications.

   In conclusion, we showed how a nontargeted metabolomics technique can
   effectively distinguish between ischemic stroke and HCs and link
   various metabolites to phenotypes of illness or well state. Moreover,
   six fecal metabolites are promising biomarkers for differentiating
   ischemic stroke from HCs, awaiting additional validation studies
   conducted over an extended period in bigger cohorts across ethnic
   landscapes. Additionally, we discovered that these six metabolites
   together could produce very good diagnostic results in both training
   and testing sets. While these metabolites show promise as acute-phase
   indicators, their role in pre-stroke risk prediction requires further
   investigation. Despite certain limitations, this work provides new
   opportunities to investigate the gut metabolome in search of
   biomarkers.

Supplementary Information

   [128]Supplementary Material 1.^ (264.2KB, docx)
   [129]Supplementary Material 2.^ (43.9KB, xlsx)

Acknowledgements