Integrative metagenomic and metabolomic analyses reveal the role of gut microbiota in antibody-mediated renal allograft rejection

Totally, 60 kidney transplantation recipients from Henan Provincial People’s Hospital affiliated to Zhengzhou University were enrolled in this study, 28 of which showed AMR (AMR group) and 32 of which were with stable post-transplant renal functions (control group). This study was performed according to ethical guidelines of Henan Provincial People’s Hospital affiliated to Zhengzhou University. AMR was diagnosed with the Banff 2019 criteria [13]. Recipients were excluded if there was a recent history of infection, non-infectious diarrhea, antibiotic usage, or gastric/colon resection. Patients were asked to provide the fecal samples within 24 h after AMR diagnosis. Fecal samples from kidney transplantation recipients with stable renal functions were collected as controls. Fresh stool samples collected from each subject were immediately frozen at − 80 °C until they were processed.

About 100 mg of fecal content were used for DNA extraction using the DNeasy PowerSoil Kit (QIAGEN, Netherlands) following manufacturer’s instructions. The quantity and quality of extracted DNA were checked with a NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). Metagenome shotgun libraries with insert sizes of 400 bp were constructed for Illumina sequencers using a TruSeq Nano DNA LT Library Preparation Kit (Illumina) based on manufacturer’s protocols. Sequencing of 2 × 150-bp paired-end reads was performed on an Illumina HiSeq X-ten platform (Illumina, USA) at Personal Biotechnology Co., Ltd. (Shanghai, China).

To obtain high-quality reads for further analysis, raw reads were firstly processed using Cutadapt (v1.2.1) to trim sequencing adapters and low-quality bases, and then mapped to the host genome using BWA (http://​bio-bwa.​sourceforge.​net/​) to remove host contamination [14]. The quality-filtered reads were de novo assembled to construct the metagenome for each sample by MEGAHIT (https://​hku-bal.​github.​io/​megabox/​) [15]. The metagenomic scaffolds longer than 200 bp were used for ORF prediction by MetaGeneMark (http://​exon.​gatech.​edu/​GeneMark/​metagenome) [16]. All ORFs were clustered by CD-HIT to construct a non-redundant gene catalog (identity > 90%) [17].

Gene abundance in each sample was estimated by soap.coverage (http://​soap.​genomics.​org.​cn/​) based on the number of aligned reads. For taxa analysis, genes were searched with the lowest common ancestor (LCA) approach against NCBI-NT database by BLASTN (e value < 0.001). The abundance of a taxonomic group was calculated by summing its matching genes. For functional annotation, gene catalogs were annotated using DIAMOND against KEGG databases [18]. Antibiotic resistance and virulence genes of microbiota were identified using Antibiotic Resistance Database (CARD) and Virulence Factor Database (VFDB), respectively [19, 20].

About 100 mg of fecal sample was used for metabolite extraction. Subsequently, 1 mL of ice-cold chloroform/methanol/water (1:2:1, v/v/v) was added to each fecal sample. The homogenates were then incubated at 4 °C for 2 h and centrifuged at 13,000 rpm at 4 °C for 15 min. The supernatant was further filtered using a 0.22 μm membrane filter, blown dry with nitrogen and stored at − 80 °C until use. Meanwhile, an equal aliquot from each sample was mixed to prepare the quality control (QC) samples. A solution of isopropanol/acetonitrile/water (1:1:2, v/v/v) was used to reconstitute samples before LC/MS analysis.

LC/MS Chromatographic separation was performed on an ultra-high-performance liquid chromatography (UHPLC) DIONEX UltiMate_3000 system (Thermo Fisher Scientific, San Jose, CA, USA) equipped with a C18 column, 1.7 µm, 2.1 × 100 mm (Waters Corp., Milford, MA, USA). The flow rate was 0.35 mL/min, the injection volume was 3 μL, and the column temperature was 45 °C. Mobile Phase A consisted of water with 0.1% formic acid (FA), and Mobile Phase B consisted of acetonitrile with 0.1% FA. The gradient elution used was started from 98% A for 0.5 min, linearly decreased to 2% A for 14.5 min, held for 3 min, and finally linearly increased to 98% A to re-equilibrate for 3 min. The QC samples were inserted into the analytical queue to monitor and evaluate the system stability and data reliability. Samples were analyzed by liquid chromatography-tandem mass spectrometry (LC/MS) using UHPLC coupled to a QExactive mass spectrometer (Thermo Fisher, Bremen, Germany). Electrospray ionization (ESI) was performed in positive and negative ion modes. The conditions of the ESI source were as follows: spray voltage, 3.5 kV (ESI+) or 3.2 kV (ESI−); source temperature, 320 °C; sheath gas flow rate, 45 Arb; aux gas flow rate, 15 Arb; mass range, 80–1200 m/z; full ms resolution, 70,000; MS/MS resolution, 17,500; TopN, 10; stepped NCE, (20,40,60); duty cycle, ~ 1.2 s.

Peak alignment, retention time correction, and peak area extraction were performed using the Compound Discoverer 3.0 program [21]. Accurate mass number matching (< 25 ppm) and second-level spectrogram matching were used to retrieve the MZcloud database [11]. Orthogonal partial least-squares-discriminant analysis (OPLS-DA) analysis was performed using the Pareto scaling method and SIMCA-P software [22]. Metabolites with both multidimensional statistical analysis VIP > 1 and univariate statistical analysis P value < 0.05 were selected as metabolites with significant differences. The sample preparation and subsequent metabolomic analysis were conducted at Shanghai Personal Biotechnology Co., Ltd. (Shanghai, China).

Alpha-diversity indices (ACE, Chao1, Shannon, and Simpson) were calculated with QIIME (Version 1.9.0). The statistical significance of alpha diversity between groups was evaluated by Mann Whitney U test or Student’s t-test using SPSS. Beta-diversity was calculated by nonmetric multidimensional scaling (NMDS) and hierarchical clustering with the QIIME. Differential abundance of taxa, KO and metabolites was tested by Wilcoxon rank sum test. Only species or KOs with an average relative abundance above 10⁻⁷ were considered in the analyses. Linear discriminant analysis effect size (LEfSe) was also utilized to compare and visualize significant differences in species between groups [23]. Receiver operating characteristic (ROC) analysis was performed to evaluate the diagnostic value.

A total of 60 kidney transplantation recipients, including 28 individuals with AMR and 32 non-rejection controls. Demographic information and clinical characteristics of the recipients were provided in Table 1. The histopathological characteristics of renal biopsy samples for the AMR cases scored according to the Banff 2019 criteria were shown in Additional file 1: Table S1. No differences in age, gender or BMI were detected between the two groups (P > 0.05; Table 1). Significantly higher levels of serum creatinine (Cr, P < 0.0001), blood urea nitrogen (BUN, P < 0.0001), uric acid (UA, P = 0.0198), serum cystatin C (CysC, P < 0.0001), serum C-reaction protein (CRP, P = 0.0005), and urine protein (P < 0.0001), and lower levels of serum carbon dioxide (CO₂, P = 0.0019), serum albumin (ALB, P < 0.0001), total bile acid (TBA, P = 0.0257), hemoglobin (HGB, P < 0.0001), and white blood cell (WBC, P = 0.0091) were observed in recipients with AMR.

To comprehensively explore gut microbiota linked to AMR after kidney transplantation, metagenomic sequencing of fecal samples from the AMR and control groups were performed on the Illumina HiSeq X-ten platform at an average depth of about 83 M reads (13 G bp) per sample. We first examined the differences in gut microbial alpha diversity between the AMR and control groups. A significant decrease in species richness (Chao1: P = 0.0055; ACE: P = 0.0038) was detected in the AMR group, while no differences in microbiota community diversity (Simpson: P > 0.05; Shannon: P > 0.05) were observed between the two groups (Fig. 1A). For beta diversity, NMDS using Bray–Curtis distances revealed a different distribution of gut microbiota between the AMR and control groups (P = 0.002, ANOSIM, Fig. 1B). The relative proportion of dominant taxa at the phylum and genera levels, and their contribution to each group were shown in Fig. 1C. To dissect the detailed taxonomic features involved in AMR, we analyzed metagenomics data at species level. A total of 5554 species were annotated, including 300 species unique to the AMR group, 124 unique to the control group, and 5130 shared by both groups (Fig. 1D). Comparisons between groups were performed using nonparametric analysis. There were 292 and 19 species showing significant decrease and elevation (fold change > 1.5 and P < 0.05), respectively (Fig. 1E and Additional file 1: Table S2). The differential species were mainly from Proteobacteria, Actinobacteria, Firmicutes, and Bacteroidetes phyla, and about 40% of the species elevated in the AMR group were from Firmicutes phylum. The differential abundances of gut microbial taxa at the phylum and genus levels were presented in Additional file 1: Tables S3 and S4. LEfSe analysis was also performed to identify the specific species associated with AMR. The results showed elevated Klebsiella phage KP8, Lactobacillus fermentum, Enterococcus phage IME-EFm1, and Streptococcus sp. I-P16, and 30 decreased species including Roseburia intestinalis, [Eubacterium] rectale and Blautia obeum in the AMR group (Fig. 1F).

To investigate the functional properties of gut microbiota between the AMR and control groups, we next performed functional annotations of the metagenome to KEGG modules. Totally, we obtained 8,217 KEGG orthologs (KOs), and most of them were related to metabolism (Fig. 2A). Compared to the control group, 213 KOs were significantly up-regulated, and 224 were significantly down-regulated in the AMR group (Fig. 2B and Additional file 1: Table S5). The first 50 differential KOs of average abundance were exhibited in the heatmap (Fig. 2C). Next, the differentially expressed genes were subjected to KEGG functional enrichment analysis. The bubble chart showed that differential genes were mainly enriched in 22 pathways (for example, Phosphotransferase system, Limonene and pinene degradation, Flagellar assembly, Bacterial chemotaxis, Ascorbate and aldarate metabolism), 13 of which were related to metabolism (Fig. 2D). These functional shifts of microbial metagenome indicated a correlation between AMR and an imbalance of gut microbes involved in the metabolism.

Considering that the metabolic functions of gut microbiota between AMR and controls were distinct, metabolic profiling of fecal samples was further performed to assess the impact of a shifted gut microbiome on the metabolic products. The OPLS-DA score map showed a clear separation between AMR and controls using untargeted LC–MS metabolomics in positive and negative mode (Fig. 3A). A total of 8120 m/z features (4518 in POS and 3602 in NEG) were detected, among which 265 (153 in POS and 112 in NEG) were significantly up-regulated, and 607 (295 in POS and 312 in NEG) were significantly down-regulated in the AMR group (Fig. 3B). Of these, 32 differential features were successfully annotated as known metabolites (Fig. 3C). There were 11 metabolites (taurocholate, phenol, l-glutamine, alpha-ketoglutarate, N1-methyl-2-pyridone-5-carboxamide, etc.) up-regulated, and 21 metabolites (N-acetyl-l-histidine, ferulic acid, 3b-hydroxy-5-cholenoic acid, 2-isopropylmalic acid, N6, N6, N6-trimethyl-l-lysine, etc.) down-regulated in the AMR group (Fig. 3C). The KEGG analysis indicated that the differential metabolites were enriched in 20 pathways including GABAergic synapse, d-Glutamine and d-glutamate metabolism, Proximal tubule bicarbonate reclamation, Taurine and hypotaurine metabolism, and the secondary metabolite biosynthesis and metabolic pathway (Fig. 3D).

Spearman correlation analysis was conducted to further explore the relationships among the AMR-associated gut microbial species, functions and metabolites. A total of 77 microbial species were significantly correlated with 16 functional genes, which were further correlated with 4 metabolites (P < 0.05, r > 0.5 or < − 0.5; Fig. 4 and Additional file 1: Table S6). Due to space limitations, the Sankey plot in Fig. 4 presented correlations among 24 representative microbial species, functional genes and metabolites in a schematic manner. Detailed information on the correlation coefficients can be found in Additional file 1: Table S6. Most of the 16 functional genes (6-phosphogluconate dehydrogenase, glucose-6-phosphate 1-dehydrogenase, indolepyruvate ferredoxin oxidoreductase alpha subunit, glutamyl-tRNA synthetase, etc.) correlated with both microbial species and metabolites were metabolic enzyme-related genes (Fig. 4). Notably, all the 77 microbial species demonstrated in Additional file 1: Table S6 were also directly correlated to 4 metabolites including 3b-hydroxy-5-cholenoic acid, l-pipecolic acid, taurocholate, and 6k-PGF1alpha-d4. These data demonstrated the direct interaction between fecal microbiota and metabolites.

Based on our results (Table 1), 11 clinical indicators (Cr, BUN, CRP, U-Pro, etc.) were validated to be significantly different between the AMR and control groups. We then evaluated if altered gut microbial species and metabolites in the AMR group were significantly correlated with these clinical indicators using Spearman analysis. A total of 340 microbial species showed significant correlation with one or more clinical indicators (P < 0.05, r > 0.3 or < − 0.3; Additional file 1: Table S7). The top 40 species sorted by correlation relevance according to Spearman analysis were shown in Fig. 5A. Erysipelotrichaceae bacterium I46 was found to be positively correlated with CysC, BUN, Cr, U-Pro and CRP, and negatively correlated with TBA, WBC and ALB (Fig. 5A and Additional file 1: Table S7). Differently, species such as Corynebacterium glutamicum, Martelella mediterranea, and Bifidobacterium angulatum showed positive correlation with CysC, BUN, Cr, U-Pro and CRP, and showed negative correlation with WBC, CO₂, HGB and ALB (Fig. 5A and Additional file 1: Table S7). Apart from pyridoxine, Gly-Arg, Dioscin, ferulic acid, l-glutamine and taurochola, all the fecal metabolites altered in AMR shared some degree of relatedness with at least one clinical indicators (Fig. 5B and Additional file 1: Table S8). Remarkably, N1-methyl-2-pyridone-5-carboxamide and aminopterin exhibited correlation with multiple clinical indicators (Fig. 5B and Additional file 1: Table S8).

To assess the diagnostic potential of the differential gut microbiota and metabolites for discrimination between patients with AMR and controls, ROC curves were constructed. First, we performed AUC estimations based on the abundance data of the top 50 species in the average abundance. Among these species, 20 had AUC values larger than 0.7 (Additional file 1: Table S9), including 7 species with AUC values larger than 0.79, of which the ROC curves were shown in Fig. 6A. Antibiotic resistance and virulence genes of the 20 key species associated with AMR (AUC > 0.7) identified against CARD and VFDB, were shown in Additional file 1: Table S10. The AUC values of the 32 differential metabolites were also calculated, and 15 of them had AUC values larger than 0.7 (Additional file 1: Table S11). ROC curves of the metabolites including alpha-ketoglutarate, N1-methyl-2-pyridone-5-carboxamide, 2-isopropylmalic acid and 3b-hydroxy-5-cholenoic acid were shown in Fig. 6B. For most biomarkers, AUC were < 0.8 (Additional file 1: Tables S9 and S11), indicating that they had moderate diagnostic values. Thus, we analyzed the microbial and metabolic biomarkers with relatively high (Fig. 6A and B) using multivariate logistic regression. Taken together, the 7 species had a combined AUC of 0.9453, and the 4 metabolites had a combined AUC of 0.8526 (Fig. 6C). The combination model with both the microbial and metabolic biomarkers (AUC = 0.9726) outperformed the species or metabolite only model in the discrimination of the patients with AMR from the controls (Fig. 6C). These results indicated that the gut microbiota and metabolites may function as biomarkers to distinguish patients with AMR from the controls.