1. Participants

From November 2023 to August 2024, the subjects of the study included 36 LT patients in hospital from the First People's Hospital of Kunming City, who were enrolled in this study after signing the informed consent form. All patients had received tacrolimus-based immunosuppressive therapy after LT. Gut fecal and whole blood samples were collected from LT patients receiving different doses of drug treatment and at different postoperative time points. The Ethics Committee of the First People's Hospital of Kunming City & Calmette Affiliated Hospital of Kunming Medical University (No. YLS2023−50), Written informed consent was obtained from all individual participants.

Inclusion criteria were defined as: (1) Patients who underwent orthotopic LT; (2) Received tacrolimus-based immunosuppressive therapy after transplantation; (3) Voluntarily participated in the study and signed the informed consent form. Exclusion criteria were: (1) Complicated with other organ transplantation; (2) Severe infections (e.g., sepsis, severe pneumonia) within 1 month before sample collection; (3) Received probiotic, prebiotic, or antibiotic treatment within 2 weeks before sample collection; (4) Complicated with severe gastrointestinal diseases (e.g., inflammatory bowel disease, intestinal obstruction); (5) Pregnancy or lactation; (6) Mental disorders or inability to cooperate with sample collection and clinical data documentation.

Fecal and whole blood samples were collected from all participants, who were stratified based on different postoperative time points and tacrolimus dosage regimens. Key clinical variables, including patient age, surgical technique (conventional vs. piggyback), type of biliary reconstruction, immunosuppressant regimens, and antibiotic usage, were documented.

2. Sample collection

Fecal samples were collected once from each participant at least 1 month after LT and when they were in a stable clinical state (no active infection or rejection): participants were instructed to urinate first to avoid contamination, defecate into a sterile container, immediately use the sterile spoon attached to the sampling tube to collect 3–5 g of the middle section of feces (to minimize air exposure), place the collected feces and the spoon together into a cryogenic tube containing 5 mL of fecal preservation solution, seal the tube tightly and label it with the participant's ID and collection time, temporarily store it at −20°C (for no more than 2 hours) before transferring it to −80°C for long-term storage, and ship the sample to the laboratory on dry ice within 1 week.

After the participant had fasted for 8–12 hours, 5 mL of antecubital venous blood was collected using a vacuum blood collection tube containing EDTA-K2. The tube was immediately gently inverted 5–8 times to mix the blood (avoiding vigorous shaking to prevent hemolysis). Within 30 minutes of collection, the mixed whole blood was aliquoted into pre-labeled cryogenic tubes using a sterile pipette (recommended aliquot volume: 500 μL to 1 mL per tube), pending tacrolimus whole blood concentration testing and subsequent analysis. Detailed records of the sample collection time, sample ID, processing procedures, and any abnormal conditions were maintained throughout the entire process.

3. Analytical methods

1. (1). Tacrolimus Blood Concentration Monitoring: Blood concentration monitoring of tacrolimus was performed in patients who had received tacrolimus for at least 3 days. In the early morning under fasting conditions, 2 mL of venous blood was collected and placed into EDTA-K2 vacuum tubes for subsequent analysis. Tacrolimus concentrations were measured using the i1000 analyzer with the corresponding reagent kits, strictly following the standard operating procedures of the i1000. Internal quality control was implemented throughout the assays.
2. (2). Metagenomic sequencing: Microbial DNA was extracted from fecal samples using the E.Z.N.A.® stool DNA Kit (Omega Bio-tek, Norcross, GA, U.S.), following the manufacturer's instructions. Metagenomic sequencing libraries were constructed and sequenced at Shanghai Biozeron Company. The process is briefly described as follows: 1 μg of DNA was fragmented into approximately 450 bp segments using a Covaris S220 ultrasonic machine (Woburn, MA USA) and then subjected to PE150 sequencing on the Illumina HiSeq X platform. Quality control was performed using Trimmomatic (https://github.com/usadellab/Trimmomatic) software to remove low-quality and contaminated reads. The reads that passed quality control were mapped to the human genome (version: hg19) using the BWA mem algorithm to remove host contamination and obtain clean reads [22]. Bacterial annotation was performed by aligning with the UHGG database [23], and viral annotation was done by aligning with the MGV database [24]. The BLASTN tool was used to compare with the UHGG database to identify and quantify the microbial composition.
3. (3). Collect intestinal microbiota samples and immediately place them into sterile EP tubes, with complete labeling of information and shortening the interval between collection and processing to maintain the integrity of the microbiota. Add 80% methanol aqueous solution to each EP tube, operate carefully to avoid splashing, and ensure thorough mixing. Subsequently, immerse the EP tubes in liquid nitrogen for quick-freezing for 3–5 minutes to terminate reactions and maintain the integrity of analytes; after removal, thaw them at room temperature or in a water bath with a temperature ≤ 37°C. Once thawed, vigorously vortex for 1–2 minutes to disrupt cell membranes, and then use an ultrasonic cleaner or probe-type ultrasonic disruptor for ultrasonic treatment for 5–10 minutes to enhance extraction. Transfer the treated samples to centrifuge tubes, centrifuge at 10,000–15,000 rpm at 4°C for 10–15 minutes, and carefully collect the supernatant containing the target analytes. Transfer the supernatant to a suitable container, and freeze-dry it at 0.01–0.05 mbar and −50 to −80°C for 8–12 hours until it becomes a dry powder. Add 100–200 μL of methanol to the dry residue and vortex to dissolve [24]. Finally, load the samples into the Vanquish UHPLC system (Thermo Fisher, Germany) coupled with the Orbitrap Q Exactive™ HF mass spectrometer (Thermo Fisher, Germany) at Biozeron Co., Ltd. (Shanghai, China), set the chromatographic and mass spectrometric parameters according to the characteristics of the target analytes, and perform three or more parallel analyses to ensure the reproducibility of the results [25].
4. (4). Metabolite Identification and Quantification: The raw data files were processed using the CD3.1 library search software, where each metabolite was subjected to a preliminary screening based on parameters such as retention time and mass-to-charge ratio. Subsequently, relevant information was set for peak extraction, and metabolite identification was performed by comparing the high-resolution tandem mass spectra with databases including mzCloud, mzVault, and MassList. The identified metabolites were statistically categorized according to the ClassyFire (http://classyfire.wishartlab.com/) classification information. Functional and categorical annotations of the identified metabolites were conducted using databases such as the Kyoto Encyclopedia of Genes and Genomes (KEGG, https://www.genome.jp/kegg/pathway.html), the Human Metabolome Database (HMDB, https://hmdb.ca/metabolites), and LIPID MAPS (https://www.lipidmaps.org/). Chromatographic peak integration was carried out for the detected peaks in the samples, with background ions removed using blank samples. The raw quantitative results were normalized using the total peak area. Metabolites with a Coefficient of Variation (CV) of less than 30% in the Quality Control (QC) samples were retained for the final identification and relative quantification results of the metabolites.

4. Statistical analysis

1. (1). Diversity Analysis: The abundance and diversity of the microbial community can be reflected by single-sample diversity analysis (Alpha diversity). The mothur indices (version v.1.30.1; http://www.mothur.org/wiki/Schloss_SOP#Alpha_diversity) were used to calculate the Shannon, ACE, Chao1, and Simpson indices. The Chao index is used to estimate species richness, the Shannon index and Simpson index calculate species diversity, and the ACE index is an index that estimates species diversity based on rare species; a higher value indicates a greater variety of species in the community. Using R language (https://www.r-project.org/) to plot a boxplot that displays the differences in diversity indices under different treatments. The differences between the two groups were assessed using the Wilcoxon rank-sum test (P ≤ 0.05). The variation among the three groups was evaluated using the Tukey’s HSD test (using the “multcomp” package, P ≤ 0.05). Statistical analysis was performed on the abundance of bacteria and viruses at different taxonomic levels (phylum, class, order, family, genus, and species). The degree of difference in sample species abundance distribution can be quantified through statistical distances. The Bray-Curtis algorithm was used to calculate the distance between sample pairs, resulting in a distance matrix. QIIME (https://qiime.org/) was used to compute the beta diversity distance matrix. R language was employed for principal co-ordinates analysis (PCoA) statistical analysis and plotting to study the similarity of sample community composition. Non-parametric multivariate statistical tests (Adonis) were performed using R vegan package to assess the statistically significant differences of bacterial community diversity indices between samples. Differences were considered significant at p < 0.05.
2. (2). Variation Analysis: Before the analysis, since many species were detected only in individual samples, especially viruses, to reduce false positives, species with a detection rate of less than 60% in the study samples were first removed. Subsequently, the Wilcox rank-sum test (p value ≤ 0.05), random forest, and LEfSE (https://huttenhower.sph.harvard.edu/lefse/) analysis were used to identify potential key species that could distinguish between groups at various levels of bacteria and viruses, respectively, to screen for usable species for subsequent analyses. In the functional gene analysis, the Wilcox rank-sum test and random forest analysis were employed to identify differentially expressed functional genes. To ensure data quality and comparability, QC samples were used to monitor system stability. Signal drift was corrected using QC-based LOESS normalisation, followed by total ion current (TIC) normalisation. Features exhibiting poor reproducibility in QC samples (RSD > 30%) were excluded from subsequent analyses. Based on the metabolite abundance results of each sample, the metabolic profiles of positive and negative ion identifications for grouped samples were combined, and Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA) was used to filter out orthogonal signals that were not related to model classification. The permutation test was performed on OPLS-DA (MetaboAnalyst, https://github.com/xia-lab/MetaboAnalystR). We applied univariate analysis (t-test) to calculate the statistical significance (P-value). The metabolites with P-value< 0.05 and fold change ≥2 or FC ≤ 0.5 were considered to be differential metabolites. Volcano plots were used to filter metabolites of interest which based on log2 (Fold Change) and -log10 (p-value) of metabolites by ggplot2 in R language. The lollipop chart screens were used to filter metabolites of interest which based on P-value values tested by VIP and T-test. Association statistical analysis and figures were performed using the Biozeron Cloud Platform [26].
3. (3). Functional Gene Study: Based on the annotation results from the UHGG and MGV databases, which correspond to Clusters of Orthologous Groups of proteins (COG) and the Kyoto Encyclopedia of Genes and Genomes (KEGG), proteins were functionally classified according to COG and KEGG annotations to obtain the functional abundance of COG and KEGG. According to sample grouping, potential functional genes that could distinguish between groups were identified at the COG and KEGG L1, L2, and L3 levels, and the viable functional genes were selected for subsequent analysis.

1. Associations between tacrolimus exposure and gut microbiota composition in liver transplant recipients

To explore the associations between tacrolimus exposure and gut microbiota of LT recipients, we collected fecal samples from 36 LT patients who received tacrolimus-based immunosuppressive therapy postoperatively. The clinical characteristics of these patients are summarized in Table 1. Patients were grouped based on the duration of tacrolimus use: the short-term use group (ST group, n = 21, ≤ 3 months) and the long-term use group (LT group, n = 15, > 3 months). As the core immunosuppressant in LT therapy, tacrolimus has a narrow therapeutic window. Direct measurement of whole blood tacrolimus concentration can reflect the actual exposure level of the drug in the systemic circulation of recipients at the time of sampling. According to the tacrolimus concentration in whole blood samples, the patients were further divided into three groups: the low concentration group (LC group, n = 19, whole blood tacrolimus concentration ≤5 ng/ml), the medium concentration group (MC group, n = 14, whole blood tacrolimus concentration 5–10 ng/ml), and the high concentration group (HC group, n = 3, whole blood tacrolimus concentration ≥10 ng/ml). A key limitation of analyzing concentration alone is the inter-individual variability in drug metabolism. By grouping patients based on the concentration-dose ratio, the impact of individual metabolic capacity on the gut microbiota can be isolated. Individuals were grouped based on the drug concentration-to-dose ratio (drug dose per kilogram of body weight, C/D): those with a C/D ratio of 50–100 were classified into the low C/D ratio group (LCD group, n = 7), and those with a C/D ratio >100 were classified into the high C/D ratio group (HCD group, n = 17). The remaining samples were excluded due to the lack of patient weight data or a C/D ratio <50 (Fig 1). The cases with a concentration-dose ratio <50 are mostly attributed to abnormally high individual metabolic efficiency in patients, which may be accompanied by other confounding factors. These factors could independently affect the gut microbiota and metabolites, rather than being driven solely by the C/D ratio.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Clinical characteristics of liver transplant recipients.

https://doi.org/10.1371/journal.pone.0343817.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Experimental design for stratification of tacrolimus-treated liver transplant patients and multi-omics sample analyses.

Fig 1 illustrates the experimental design where liver transplant patients taking tacrolimus were divided into ST and LT groups based on medication duration, and into LC, MC, and HC groups (N = 3) based on the drug concentration of tacrolimus in whole blood, as well as into LCD and HCD groups based on the drug concentration-to-dose ratio. Fecal samples were collected for metagenomic testing, and whole blood samples were analyzed for metabolomics.

https://doi.org/10.1371/journal.pone.0343817.g001

The overall composition of the gut microbiota in the study cohort was first characterized. At the bacterial genus level, the top 10 most abundant taxa in post-treatment LT patients were Bacteroides, Phocaeicola, Escherichia, Parabacteroides, Enterocloster, Veillonella, Alistipes, Klebsiella_A, Klebsiella, and Prevotella (Fig 2A). At the species level, the top 10 abundant bacterial species in LT patients after medication were Phocaeicola dorei, Escherichia coli_D, Bacteroides fragilis, Bacteroides stercoris, Bacteroides uniformis, Bacteroides thetaiotaomicron, Klebsiella_A michiganensis, Phocaeicola plebeius_A, Bacteroides xylanisolvens, and Bacteroides ovatus (Fig 2D). At the viral genus level, the top 10 abundant viral genus in LT patients after medication were Brunovirus, Uetakevirus, Peduovirus, Lubbockvirus, Punavirus, Drulisvirus, Lederbergvirus, Svunavirus, Bcepmuvirus, and Teseptimavirus (Fig 2G).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Relative abundances of dominant bacteria and viruses in tacrolimus-treated LT patient subgroups.

Panel A, Panel D, Panel G show the bar graphs of the relative abundance of dominant bacteria at the genus and species levels and dominant viruses at the genus level for the ST and LT groups. Panel B, Panel E, Panel H show the bar graphs of the relative abundance of dominant bacteria at the genus and species levels and dominant viruses at the genus level for the LC, MC, and HC (N = 3) groups. Panel C, Panel F, Panel I show the bar graphs of the relative abundance of dominant bacteria at the genus and species levels and dominant viruses at the genus level for the LCD and HCD groups.

https://doi.org/10.1371/journal.pone.0343817.g002

Intergroup comparisons suggested a trend toward associations between tacrolimus treatment duration and gut microbial composition. Relative to the ST group (≤3 months of tacrolimus use), the LT group (>3 months of tacrolimus use) exhibited increased abundance of Bacteroides and Escherichia and decreased abundance of Phocaeicola at the bacterial genus level (Fig 2A). At the species level, the abundance of Escherichia coli_D and Bacteroides stercoris increased, while the abundance of Phocaeicola dorei and Bacteroides fragilisdecreased (Fig 2D). At the viral genus level, the abundance of Brunovirus and Uetakevirus decreased (Fig 2G). Comparisons based on different whole blood concentrations of tacrolimus found that at the bacterial genus level, with increasing whole blood drug concentrations, the abundance of Bacteroides and Escherichia gradually decreased, while the abundance of Klebsiella increased (Fig 2B). At the species level, Klebsiella varicola and Bacteroides fragilis_A had increased abundance in the HC (N = 3) group and were negligible in the MC group and LC group, with Phocaeicola dorei being most abundant in the MC group (Fig 2E). At the viral genus level, the abundance of Peduovirus increased in the HC (N = 3) group, while Brunovirus and Uetakevirus were relatively more abundant in the MC group (Fig 2H).

In the C/D ratio stratification, higher C/D ratios were significantly associated with alterations in gut microbial composition. At the bacterial genus level, the abundance of Bacteroides increased with increasing C/D ratio, while the abundance of Escherichia and Parabacteroides decreased (Fig 2C). At the species level, the abundance of Bacteroides fragilis and Escherichia coli D decreased in the HCD group (Fig 2F). At the viral genus level, the abundance of Peduovirus increased in the HCD group (Fig 2I).

2. Associations between tacrolimus exposure and gut microbial diversity

To compare the richness and evenness of gut microbiota among the groups, four different indices (Chao1 index, ACE index, Shannon index, and Simpson index) were used to compare across different groups. The comparison results showed that the species level of bacteria, the index values of the LT group were all higher than those of the ST group, with Chao1 index (p = 0.038), ACE index (p = 0.040), Shannon index (p = 0.017), and Simpson index (p = 0.042) (Fig 3A). At the species level of viruses, the index values of the LT group were all higher than those of the ST group. There were significant differences between the ST and LT groups for the Chao1 index (p = 0.014), ACE index (p = 0.013), and Shannon index (p = 0.022). However, for the Simpson index (p = 0.764), indicating no statistically significant difference (Fig 3B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Gut bacterial and viral α-diversity comparisons across tacrolimus-treated liver transplant patient subgroups.

Panel A presents a comparison of the gut bacterial α-diversity indices between the ST and LT groups across four different indices: the Chao1 index, ACE index, Shannon diversity index, and Simpson index. Each subplot features box plots that show the distribution range, median, and quartiles of the data for each group, with p-values less than 0.05 on each box indicating significant differences between groups. Panel B shows the comparison of gut viral α-diversity indices between the ST and LT groups across the same four indices. Panel C-F respectively display the Chao1 index and Shannon diversity index for bacteria and viruses in the LC, MC, and HC (N  =  3) groups, as well as the LCD and HCD groups. Note: The “a” labels in the figure are formatting artifacts and can be ignored.

https://doi.org/10.1371/journal.pone.0343817.g003

In contrast, no significant differences in bacterial and viral diversity among the different tacrolimus blood concentration groups across the indices (all p > 0.05) (Fig 3C-3D). Similarly, no significant differences in bacterial or viral diversity were found between the LCD and HCD groups (all p > 0.05) (Fig 3E-3F).

To further explore the diversity differences of gut microbiota among the groups, we performed PCoA analysis. In the drug administration time groups, the ST and LT groups showed almost complete overlap in bacterial communities, with high similarity between the two groups. For viral communities, there was a high degree of overlap, indicating similar community structures (Fig 4A). In the different blood drug concentration subgroups, the bacterial communities in the gut mostly overlapped, with high similarity across all groups. The viral communities showed large overlap in the middle part, with relatively high similarity, but the distribution trends varied, and there were some differences with changes in concentration (Fig 4B). In the different C/D ratio subgroups, we found that the bacterial communities in the LCD and HCD groups were almost identical, and the viral communities completely overlapped (Fig 4C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. PCoA Plots and PERMANOVA (Adonis) analysis of gut microbiota dissimilarities in stratified patient cohorts.

Panel A-C present the PCoA analysis for the different groups, reflecting the differences in the data on a two-dimensional coordinate plot, with the axes representing the two principal components that maximize the variance. Samples with more similar compositions are closer together on the PCoA plot. The results of the PERMANOVA (Adonis) analysis (R² and p-values) are shown in the figure.

https://doi.org/10.1371/journal.pone.0343817.g004

3. Associations Between tacrolimus exposure and discriminatory gut microbiota taxa, and microbial functional gene profiles

Based on the sample grouping, potential key discriminatory taxa that distinguish intergroup differences were identified. In the medication duration-based grouping, the most significant differential bacteria at the genus level were Klebsiella_A and Enterobacter, with high abundance in the ST group. The significant differential viruses at the genus level were Cepunavirus and Teseptimavirus, with higher abundance in the LT group (Fig 5A). For the grouping by different blood drug concentrations, the most significant differential bacteria CAG-306 and the significant differential virus Rauchvirus were of the highest importance in the respective groups (Fig 5B). In the grouping by different C/D ratios, the primary differential bacteria at the genus level were Enterococcus_B with high abundance in the LCD group, and the main significant differential virus was Brunovirus with the highest abundance in the HCD group (Fig 5C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Identification and analysis of key biomarkers, functional traits, and their abundance changes across patient subgroups.

Panel A, C are combination plots of biomarker abundance changes between groups as determined by the wilcox rank-sum test. The left bar graphs show the average abundance of identified key species in the study groups, with the X-axis representing the average proportion and the Y-axis listing different bacterial genera. Red squares represent Group ST, and blue squares represent Group LT. The right scatter plots show the abundance changes of key species between groups, with the X-axis displaying the difference in average proportions. The numeric value on the far right is the P-value obtained from the Wilcox rank-sum test between species groups. Panel B shows the key species in different blood drug concentration groups, with the left dot plot displaying the identified Biomarkers and their importance to the grouping, retaining results with importance higher than 1.5. The right heatmap uses a color gradient to show the abundance of Biomarkers in the study samples, with redder colors indicating higher abundance, and gray representing the absence of the Biomarker in the sample. Panel D, F list multiple features on the left, related to the duration of medication. Each feature has different red shaded areas in each group, indicating the relative importance of the feature in these two classifiers. The darker the color, the higher the importance of the feature in the classifier. Panel E is a combination plot of key functional abundance changes between groups as determined by the Wilcox rank-sum test. The left chart shows the average proportion for each process, with red representing Group ST and blue representing Group LT. The right chart shows the average difference and 95% confidence intervals for these processes between the two groups.

https://doi.org/10.1371/journal.pone.0343817.g005

To characterize the functional features of the gut microbiota, a non-redundant gene set from metagenomic sequencing data, gene functional classification and annotation were performed, and inter-group differences were analyzed. In the blood drug concentration groups, gene functions such as Fc gamma R-mediated phagocytosis, Parathyroid hormone synthesis, secretion and action, and Endocrine and other factor-regulated calcium reabsorption were of higher importance (Fig 5D). In the tacrolimus treatment duration subgroups, the chemical carcinogenesis and bacterial invasion of epithelial cells were significantly higher in the ST group than in the LT group, and the Mannose type O-glycan biosynthesis gene function was significant lower in the ST group compared with the LT group (Fig 5E). In the C/D ratio subgroups, the circadian rhythm gene function was of high importance (Fig 5F).

4. Liver transplantation recipients treated with tacrolimus exhibited differences in gut microbial metabolites

To explore the relationship between tacrolimus exposure and gut microbial metabolite patterns, metabolic profiles were compared across subgroups stratified by tacrolimus treatment duration, whole-blood concentration, and C/D ratio. All identified metabolites from the samples were classified and statistically analyzed based on the Classy Fire classification information. The analysis revealed that the most abundant class was Lipids and lipid-like molecules, accounting for 22.26%, followed by Organic acids and derivatives at 10.38% (Fig 6A). PCA analysis of the combined metabolite abundance results from each group showed that the groups with different medication durations basically overlapped, indicating similar metabolite profile between groups (Fig 6B). In the grouping by blood drug concentration, the HC group completely overlapped with the LC and MC groups, while the LC and MC groups mostly overlapped but showed different trends (Fig 6C). The groups divided by C/D ratio mostly overlapped, but the trend directions of the two groups were different (Fig 6D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Profiling of metabolite categories and differential metabolite analysis (PCA, Volcano Plots, Radar Plots) in tacrolimus-treated cohorts.

Panel A shows each color representing a metabolite classification, with unknown classifications represented as Other. The percentage indicates the proportion of metabolites belonging to this type out of all identified metabolites. Panel B-D are PCA score plots. After dimensionality reduction analysis, samples have relative coordinate points on the principal components PC1 and PC2. The distance between the coordinate points represents the degree of clustering and dispersion among samples; a closer distance indicates higher similarity between samples, while a greater distance indicates greater differences. PCA analysis can be used to observe the trend of separation between groups in the experimental model, as well as the presence of any outliers, and reflects the variability between and within groups from the original data. The confidence ellipses represent the “true” samples of the group, which are distributed within this area at a 95% confidence level; samples outside this area can be considered as potential outliers. Panel E, Panel F are volcano plots of differential metabolites for different medication duration and different drug concentration-to-dose ratio groups, respectively. These volcano plots use consistent screening thresholds: fold-change cutoff of |log₂(FC)| ≥ 1 (actual fold change ≥ 2 or ≤ 0.5) to distinguish biologically significant differences from experimental fluctuations, and independent samples t-test p < 0.05 to control the false positive rate. OPLS-DA model VIP ≥ 1 is additionally used as an auxiliary criterion to enhance screening accuracy. The color of the points in the graph represents the category of the metabolites, with red indicating significantly up-regulated metabolites that meet the threshold, blue indicating significantly down-regulated metabolites that meet the threshold, gray indicating metabolites that do not meet the fold change screening threshold, and tan indicating metabolites that meet the fold change threshold but do not meet the P-value threshold. Panel F, Panel G are radar plots of differential metabolites for different medication duration and different drug concentration-to-dose ratio groups, respectively. For each group comparison, we calculate the corresponding ratio for the quantitative values of differential metabolites and apply a logarithmic transformation with base 2, selecting the top 10 differential metabolites with significant changes (up and down regulation combined) for display. The grid lines correspond to the log₂FC. The green colored area is formed by the lines connecting the dots (metabolites).

https://doi.org/10.1371/journal.pone.0343817.g006

Subsequent screening and visualization of differential metabolites revealed distinct metabolite alterations associated with tacrolimus exposure variables. In the medication duration groups, with the extension of medication time, 16 metabolites including Astragaloside A significantly increased, while 108 metabolites including TLK, QLK, and Capsaicin significantly decreased (Fig 6E-6F); in the C/D ratio groups, with the increase of the drug C/D ratio, 57 metabolites including 11-keto Testosterone (CRM) and L-Tryptophan significantly increased, while 13 metabolites including FAHFA 2:0/16:0 significantly decreased (Fig 6G-6H).

5. Correlation analysis of the gut microbiota with faecal metabolites and clinical variables

To evaluate the associations between clinical variables on the gut microbiota and metabolome, we conducted a correlation analysis. Various clinical variables were found to be correlated with top 30 differentially accumulated metabolites (ST vs LT) to varying degrees (Fig 7A). Some metabolites showed significant correlations with BMI, height and dosing-related parameters (mg/kg, Con) (P < 0.05 or P < 0.01), whereas the correlations between age, antibiotic use and most metabolites were weaker or did not reach statistical significance. Regarding the correlations between the relative abundances of the top 30 bacterial genera (Fig 7B): Several bacterial genera showed significant correlations (P < 0.05 or P < 0.01) with multiple physical parameters (height, weight and BMI), whilst total day and antibiotic usage also exhibited significant positive or negative correlations with the abundance of certain genera. In contrast, age demonstrated weaker correlations or failed to reach statistical significance with most bacterial genera.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Spearman correlation analysis of clinical parameters, top 30 bacterial genera, and key differential metabolites.

Panel A and Panel B show heatmaps of clinical indicators and the top 30 abundant genera, alongside the Spearman correlation coefficient for top 30 differentially accumulated metabolites (ST vs LT). The color scale shows the correlation coefficients, with red representing positive correlations and blue representing negative ones. Asterisks indicate statistical significance (* P < 0.05, ** P < 0.01).

https://doi.org/10.1371/journal.pone.0343817.g007

Spearman’s rank correlation analysis was used to evaluate the associations between the predominant bacterial genera and the top 30 differentially accumulated metabolites (ST vs LT), and hierarchical clustering was performed based on the abundance of the genera. Distinct correlation patterns emerged between different genera and metabolites (Fig 8). Certain bacterial genera (e.g., Faecalibacterium, Roseburia, Blautia and Lachnospira) showed significant positive correlations with multiple metabolites, including cafestol and ethyl oleate. In contrast, several genera (e.g., Klebsiella, Enterobacter and Escherichia) predominantly displayed significant negative correlations with specific metabolites (P < 0.05 or P < 0.01) such as spermidine. Hierarchical clustering revealed that genera exhibiting similar correlation patterns clustered together.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Heatmaps showing Spearman correlations between top 30 abundant genera and top 30 differentially accumulated metabolites.

Heatmaps illustrate Spearman correlations between the top 30 abundant genera and the top 30 differentially accumulated metabolites (ST vs. LT groups). Red and blue colors on the scale represent positive and negative correlations, respectively, and asterisks indicate statistical significance (* P < 0.05, ** P < 0.01, *** P < 0.001).

https://doi.org/10.1371/journal.pone.0343817.g008