This was a prospective observational study of children admitted to the pediatric intensive care unit (PICU) of Xinhua Hospital, affiliated with the Shanghai Jiao Tong University School of Medicine, from May 2021 to May 2022. The inclusion criteria were as follows: (i) age >28 days and <18 years, and (ii) sepsis diagnosis based on the sepsis-3 criteria (Singer et al., 2016). The exclusion criteria were as follows: (i) history of kidney disease; (ii) no follow-up due to death or discharge within 48h; (iii) refusal to participate in the study by the child’s guardian; and (iv) missing baseline data ( Figure 1A ). The patients were categorized into groups based on the presence or absence of AKI during hospitalization. AKI diagnosis was using Kidney Disease: Improving Global Outcomes (KDIGO) criteria including both plasma creatinine and urine output criteria (Kellum and Lameire, 2013).

After patient enrollment, a 48h follow-up period was conducted. Stool, blood, and urine specimens were collected upon admission [day 0 (D0)], succeeding day 1 (D1) (within the subsequent 24h post admission), as well as on day 2 (D2) (approximately 24h–48h succeeding admission). Figure 1B illustrates the study design and the time points for sample collection. This study is reported in concordance with Strengthening the Reporting of Observational Studies in Epidemiology recommendations (von Elm et al., 2007).

The baseline data included age, sex, height, weight, and body mass index at admission. Clinical indicators were assessed during the first 24h in the PICU, including kidney injury markers [serum creatinine (SCr) and blood urea nitrogen (BUN)], hemodynamic parameters [mean arterial pressure (MAP), CO, fractional shortening (FS), ejection fraction (EF), and renal resistive index (RRI)], oxygen metabolism indices (pressure of oxygen in artery (PaO₂), partial pressure of carbon dioxide in artery (PaCO₂), saturation of arterial blood oxygen (SaO₂), lactic acid, pressure of oxygen in urine (PuO₂), and partial pressure of carbon dioxide in urine (PuCO₂), and inflammatory indicators (white blood cell, the neutrophil to lymphocyte ratio, IL-8, IL-1β, IL-6, TNF-α, IL-2, IL-10, and procalcitonin). The severity of the disease is evaluated based on the PCI score, PRISM III score, and site of infection. Therapeutic strategies are assessed, which include the administration of antibiotics, vasoactive drugs, mechanical ventilation, and renal replacement therapy. Additionally, the duration of hospitalization and stay in the intensive care unit are also taken into consideration.

The primary objectives were changes in the gut microbiota and blood and urine metabolites between the SA-AKI and non-AKI groups. Secondary objectives included correlations between the gut microbiota or metabolites, clinical indicators, and associations between the gut microbiota and metabolites. Raw data can be accessed from the NCBI BioProject (PRJNA1025259).

Stool microbiome samples were routinely collected on the day of admission using sterile anal swabs and stored at −80°C without preservatives before processing. The Dinfectome Company (Nanjing, China) performed metagenomics next-generation sequencing (NGS) of stool samples as follows: DNA was extracted using the TIANamp Magnetic DNA Kit (Tiangen, Beijing) per the manufacturer’s protocols. RNA was extracted from the bronchoalveolar lavage fluid, plasma, and other samples using the QIAamp Viral RNA Mini Kit (Qiagen, Germany), and a library was constructed after Qubit quantification. DNA libraries were prepared using the Hieff NGS C130P2 OnePot II DNA Library Prep Kit for MGI (Yeasen Biotechnology, Shanghai), per the manufacturer’s protocol. The rRNA was removed from the total RNA, and a library was constructed after purification. Agilent 2100 was used for quality control, and the DNA libraries were 50 bp single-end sequencing on an MGISEQ-200. An in-house bioinformatics pipeline was used for pathogen identification. Briefly, high-quality sequencing data were generated by removing low-quality reads, adapter contamination, and duplicate and short (<36 bp) reads. Human host sequences were identified by mapping to the human reference genome (hs37d5) using the bowtie2 software (version 2.2.6). Gut microbiota data were downloaded from the human gut microbiota database (GMrepo, https://gmrepo.humangut.=home), and all annotated results (Operational Taxonomic Units, OTUs) in our study were compared to public data. Alpha diversity was estimated using Shannon and Simpson indices based on the taxonomic profile of each sample. Differential relative abundance of taxonomic groups at the species level between groups was tested using the Kruskal–Wallis rank sum test (R package kruskal.test). Genera with mean relative abundances >1% and penetrance >40% were compared among all samples. Pearson Correlation coefficients between clinical indicators and the relative abundance of genera were calculated, and false discovery rate correction was used to adjust all P-values.

Venous blood samples were collected through venipuncture in sterile 2 ml of heparinized tubes, and urine samples were collected using sterile tubes (Sterisets Urine Collection Kit). These samples were stored at −20°C before being processed. Liquid chromatography–tandem mass spectrometry (MS) was performed to analyze blood and urine samples (Fujisaka et al., 2018). The profiling procedure included experimental design, quality assurance/control procedures, sampling, metabolite extraction, metabolite measurement, data processing, post-processing, and statistical analysis. Unsupervised principal component analysis (PCA) was used to examine the overall distribution of samples and the stability of the overall analysis process. We used partial least squares discriminant analysis (PLS-DA) and orthogonal PLS-DA (OPLS-DA) to evaluate the differences in metabolic profiles between the SA-AKI and non-AKI groups. Subsequently, differentially represented metabolites were determined based on the value of the variable importance in projection, OPLS-DA model first principal component (threshold >1), and P-value of the t-test (P < 0.05). Progenesis QI (version 2.3, Nonlinear Dynamics, Newcastle, UK) was used for MS analysis. PCA was performed using the R software, and the results were visualized using the ggplot2 package (3.3.6). Heatmaps for visualization were generated using the ComplexHeatmap R package (2.13.1). Pearson’s correlations were calculated to measure the correlation between metabolite levels and clinical indicators. Kyoto Encyclopedia of Genes and Genomes (KEGG) was used to annotate pathways and classify these pathways per the KEGG website (http://www.kegg.jp/) pathway hierarchy classification method.

The microbial antibiotic resistance profiles were analyzed utilizing the Comprehensive Antibiotic Resistance Database (CARD). The comparison of antibiotic resistance genes (ARGs) from gut microbiota between groups was performed using DEseq2. The statistical difference was considered significant if the adjusted P-value < 0.05 and|log2FC|> 1. Biological processes and KEGG signaling pathways were determined with Metascape.

Data were presented as mean ± SD. Comparisons between groups were performed using the nonparametric Kruskal–Wallis test. Statistical analyses were performed using IBM SPSS Statistics for Windows, version 23.0 (IBM Corp., Armonk, NY, USA) and R (version 4.2.1). Statistical significance was set at P < 0.05.

Overall, nine patients diagnosed with sepsis, four of whom manifested SA-AKI. Baseline patient characteristics are in Table 1 . The patients’ mean age was 6.1 years (range, 1–11); seven individuals (77.8%) were men. Among the enrolled participants, 66.7% of patients developed sepsis caused by respiratory tract infections. When comparing the PCI scores and PRISM III scores between the two groups, it was observed that the patients in the SA-AKI group had more severe conditions compared to the non-AKI group.

Analysis of clinical indicators in two patient groups. In terms of kidney function indicators, patients with SA-AKI show significantly higher levels of creatinine (51.2 vs. 19.3 μmol/L), along with relatively elevated BUN levels (9.3 vs. 2.8 mmol/L). Hemodynamic analysis reveals that the non-AKI group has superior cardiac function, characterized by comparatively higher MAP, CO, and EF. However, there are no significant differences observed between the two groups in terms of RRI. In terms of oxygen metabolism analysis, the SA-AKI group exhibits higher levels of PaO₂ (86.4 vs. 71.0) and PuO₂ (162.5 vs. 117.4) compared to the non-AKI group. Analysis of inflammatory markers shows higher levels of procalcitonin in the SA-AKI group, while the non-AKI group exhibits a more pronounced immune response with higher levels of IL-8, IL-1β, IL-6, and TNF-α.

The usage of antibiotics differed between the SA-AKI and non-AKI groups. Specifically, in the SA-AKI group, 50% of the patients were treated with cephalosporins. Conversely, in the non-AKI group, a combination of third-generation cephalosporins and glycopeptide antibiotics was administered to 60% of the patients. Furthermore, the prevalence of probiotic usage (Clostridium butyricum) was higher in the non-AKI group compared to the SA-AKI group (40% vs. 25%). Patients with SA-AKI had a higher proportion of vasopressor use and mechanical ventilation compared to the non-AKI group. One patient in each group underwent CRRT. The PICU and hospital length of stay were increased in patients with SA-AKI.

We performed a comparative analysis of gut microbiota between the SA-AKI group and the non-AKI group. The Venn diagram illustrated that there were 43 OTUs shared at the species level, with a total richness of 152, between the two groups. Additionally, patients in the non-AKI group exhibited a lower number of unique OTUs compared to the SA-AKI group ( Figure 2A ). We conducted a further analysis on the disparities in gut microbiota composition between the SA-AKI group and the non-AKI group. The 20 most abundant gut microbiota at taxonomic levels were chosen to generate a histogram illustrating their relative abundance ( Figure 2B ). At the species level, Enterococcus avium showed the highest abundance in the SA-AKI group, followed by Bacteroides caccae and Campylobacter hominis. In the non-AKI group, the top three species in terms of abundance were Bacteroides fragilis, Veillonella parvula, and Prevotella corporis. The Shannon and Simpson indices displayed higher values in patients diagnosed with SA-AKI as compared to those categorized in the non-AKI group (Shannon index: 2.0 ± 0.4 vs. 1.4 ± 0.6, P = 0.230; Simpson index: 0.8 ± 0.1 vs. 0.6 ± 0.2, P = 0.494, Figure 2C ).

Next, we selected the 20 most abundant taxonomic levels of gut microbiota to conduct a correlation analysis with clinical indicators. Regarding kidney function, SA-AKI patients exhibited higher levels of SCr and BUN compared to those without AKI. Additionally, we observed an increase in the expression of Campylobacter hominis in the SA-AKI group. Moreover, we discovered a positive correlation between elevated levels of Campylobacter hominis expression and markers of AKI (SCr, Corr = 0.7, P = 0.047; blood urea nitrogen, Corr = 0.8, P = 0.012, Figure 2D ). In relation to cardiovascular function, we found a negative correlation between the levels of the Bacteroidetes (specifically Bacteroides stercoris and Bacteroides vulgatus) and cardiac systolic function (as measured by EF and FS, see Figure 2D ). Notably, higher levels of the Bacteroidetes were observed in the SA-AKI group. As for the analysis pertaining to oxygen metabolism-related indicators, our study identified an increase in the expression of Rothia mucilaginosa in the SA-AKI group compared to the non-AKI group. This elevation was found to be correlated with higher levels of PaCO₂ (Corr = 0.9, P < 0.001, Figure 2D ). In relation to inflammatory markers, a notable association was found between the levels of pro-inflammatory cytokines IL-1β, IL-6, IL-8, and TNF-α, and the presence of Bifidobacterium longum, Bifidobacterium bifidum, and Clostridioides difficile. These three types of gut microorganisms exhibited a higher prevalence in the non-AKI group as compared to the SA-AKI group.

The gut microbiota analysis results indicate a significant difference in bacterial composition between the SA-AKI group and the non-AKI group, as evidenced by the distinct relative abundance of predominant taxa. Firmicutes and Bacteroidetes were the dominant phyla in both groups, while Proteobacteria emerged as the third most prevalent phylum in the SA-AKI group. Our clinical relevance analysis further confirms the association of Proteobacteria with AKI markers. Different bacterial species show certain correlations with clinical indicators.

We continued the analysis to examine how blood metabolites changed over the first 48h after admission. On D0, 34 metabolites were identified to be downregulated, while 49 metabolites were found to be upregulated ( Figure 3A i). On D1, 48 differential metabolites were selected, with 22 downregulated and 26 upregulated metabolites observed ( Figure 3A ii). On D2, 90 differential metabolites were identified, with 36 downregulated and 54 upregulated metabolites observed. ( Figure 3A iii).

The Venn diagram presented in Figure 3B shows the unique and co-differentially expressed metabolites observed in response to AKI at three distinct time points. Specifically, three metabolites, 13Z-docosenamide, Capsiamide, and N6, N6, N6-Trimethyl-L-lysine, exhibited persistence for 48h following admission. Hierarchical clustering was applied to group these metabolites ( Supplementary Figure S1 ). Consistently across all three time points, the expression levels of 13Z-docosenamide, Capsiamide, and N6, N6, N6-Trimethyl-L-lysine were found to be higher in the SA-AKI group as compared to the non-AKI group.

The relationship between expression levels of blood metabolite and clinical indicators was explored using Pearson’s correlation analysis. Heatmap analysis was performed to investigate correlations between blood metabolites and clinical indicators ( Supplementary Figure S2 ). We identified significant correlations between blood metabolites and clinical indicators: six metabolites correlated with SCr and BUN primarily on D3, 29 with EF and FS mostly on D1, and 15 with PuO₂ and PuCO₂. Additionally, 44 metabolites, mainly on D2, were linked to inflammation markers.

Figure 3C shows the correlation between clinical indicators and the overlapping blood metabolites, 13Z-docosenamide, Capsiamide, and N6, N6, N6-Trimethyl-L-lysine, at three time points. The levels of these three blood metabolites were significantly higher in the SA-AKI group compared to the non-AKI group. In the first two days, 13Z-docosenamide and Capsiamide exhibited significant positive correlations with PuO₂, while showing a negative correlation with PuCO₂. In addition, N6, N6, N6-Trimethyl-L-lysine demonstrated a significant positive correlation with SCr.

The enriched pathways of blood metabolites were analyzed using KEGG enrichment analysis. The Venn diagram the presence of two metabolic pathways, Choline metabolism in cancer and lysine degradation, across all three-time points ( Figure 3D ). It is noteworthy that N6, N6, N6-Trimethyl-L-lysine is involved in the lysine degradation pathways.

On D0, KEGG pathway analysis identified significant enrichment of 68 pathways. Among them, the NF-kappa B signaling pathway and Th1 and Th2 cell differentiation were identified as the two most enriched pathways ( Figure 3E ). The enrichment analysis conducted on D1 revealed a significant enrichment of two KEGG pathways: Phenylalanine metabolism and Choline metabolism in cancer ( Figure 3F ). On D2, four KEGG pathways were found to be significantly enriched: Central carbon metabolism in cancer, Aminoacyl-tRNA biosynthesis, D-arginine, and D-ornithine metabolism, as well as Choline metabolism in cancer ( Figure 3G ).

The findings indicate significant alterations in blood metabolites within the initial 48h, closely linked to clinical indicators. Three metabolites maintained consistent levels during this period. Additionally, two pathways from the KEGG database were persistently active throughout the first two days.

Metabolic profiling of urine showed that a total of 39 differential urinary metabolites were identified on D0 with 22 downregulated and 17 upregulated metabolites ( Figure 4A i). On D1, nine metabolites were found to be differential, comprising three downregulated and six upregulated metabolites ( Figure 4A ii). On D2, a total of 19 differential metabolites were identified, including 11 downregulated and eight upregulated metabolites ( Figure 4A iii).

The Venn diagram depicted in Figure 4B illustrates the unique and co-differentially expressed urinary metabolites at three distinct time points. The urinary metabolite, Hydroxypyruvaldehyde phosphate, was detected on both D0 and D1, whereas three urinary metabolites, Trichloroethanol glucuronide, 2-thiothiazolidine-4-carboxylic acid, and Gibberellin A86 remained present on D1 and D2. However, no urinary metabolites persisted across D0, D1, and D2.

Hierarchical clustering was performed to identify clusters ( Supplementary Figure S3 ). We have constructed heatmaps to facilitate the visual representation of the correlation coefficients that subsist between clinical indicators and urinary metabolites ( Figure 4C ). In the aspect of kidney injury, 26 (13.7%) urinary metabolites were significantly correlated with SCr and BUN, which were mainly observed on D0 (17, 43.6%). In terms of hemodynamic, three (7.7%) urinary metabolites were observed on D1 and showed increased expression in the SA-AKI group. These metabolites were positively correlated with CO but negatively correlated with RRI. Three (15.8%) urinary metabolites on D2 were also significantly associated with CO and the RRI. Regarding oxygen metabolism, nine (3.1%) urinary metabolites were found to have a significant correlation with PuCO₂ and PuO₂ levels, primarily observed on D0 (6, 15.4%). It was observed that the urinary metabolites positively correlated with PuCO₂ increased in patients with SA-AKI. In the aspect of inflammation, 10 (3.4%) urinary metabolites were significantly associated with indicators of inflammation. Among these, on D0 and D2, they increased in the SA-AKI group and were positively correlated with indicators of inflammation; however, the situation was reversed on D1. The urinary metabolite intersection of D0 and D1, Hydroxypyruvaldehyde phosphate, showed significant associations with SCr, CO, PuCO₂, PuO₂, IL-8, and IL-10 levels on D0. However, these correlations were less pronounced on D1. Gibberellin A86, the urinary metabolite intersection of D1 and D2, was found to be significantly associated with SCr and PCO₂ on both D1 and D2.

The Venn diagram in Figure 4D illustrates the intersecting information of the enriched pathways of urinary metabolites at three time points, which were analyzed using KEGG enrichment analysis. There is no overlapping KEGG pathway between D0 and D1. Between D1 and D2, there are two intersecting KEGG pathways, namely the Chemical Carcinogenesis and the Metabolism of Xenobiotics by Cytochrome P450. A sole KEGG pathway, lysine degradation, intersects between D0 and D2. Figure 4E presents the enriched pathways of urinary metabolites at three different time points.

The metabolomic analysis of urine reveals significant changes in metabolic profiles at three distinct intervals, with no urinary metabolites demonstrating persistent expression over a 48-hour period. Additionally, no KEGG pathways maintain consistent presence throughout the same timeframe. Notably, while urinary metabolites correlate with clinical indicators, this relation does not remain stable over a prolonged period.

Existing research has established a close correlation between gut microbiota and metabolites, yet it remains unclear whether this interrelationship is subject to temporal effects (Liu et al., 2022). An UpSet graph was constructed to identify common metabolites between blood metabolites and urinary metabolites at three-time points ( Figure 5A ). Our findings revealed the presence of three metabolites—Porson, Eplerenone, and Capsiamide—in both the blood and urine on D0. In contrast, there were no detectable common metabolites in the blood and urine samples on D1. However, on D2, a congruous metabolite, N6, N6, N6-Trimethyl-L-lysine, was observed in both the blood and urine metabolites.

An UpSet graph was utilized to pinpoint shared KEGG enrichment pathways between both blood and urine metabolite enrichment pathways at three different time points ( Figure 5B ). Our research identified the existence of a single KEGG enrichment pathway, specifically lysine degradation, in both blood and urine samples at D0. Contrarily, there were no observable KEGG enrichment pathways in blood and urine specimens at D1. Nonetheless, at D2, three distinct enrichment pathways, namely, lysine degradation, Histidine metabolism, and Arginine and proline metabolism, were identified in both blood and urine metabolite KEGG enrichment pathways.

We employed Pearson’s correlation analysis to explore the correlation between gut microbiota at D0 and blood metabolites at D0 (A), D1 (B), and D2 (C) ( Supplementary Figure S4 ). Additionally, we investigated the relationship between gut microbiota at D0 and urinary metabolites at D0 (A), D1 (B), and D2 (C) using the same analytical approach ( Supplementary Figure S5 ). The three blood metabolites, 13Z-Docosenamide, Capsiamide, and N6, N6, N6-Trimethyl-L-lysine, were consistently detected at three different time points. Upon comparing their relationship with the gut microbiota at D0, it was found that there was no significant correlation between the expression of 13Z-Docosanamide and Capsaicamide with the gut microbiota. However, the blood metabolite, N6, N6, N6-Trimethyl-L-lysine, exhibited a notable positive correlation with the gut microbiota, specifically Campylobacter hominis at D0 and Bacteroides caccae at D2 ( Figure 5C ). The urinary metabolite, Hydroxypyruvaldehyde phosphate, which intersects D0 and D1, demonstrated a significant positive association with the gut microbiota Bacteroides caccae on D0, and with Rothia mucilaginosa and Enterococcus avium on D1. Moreover, the intersection of urinary metabolites between D1 and D2, namely, Trichloroethanol glucuronide and 2-thiothiazolidine-4-carboxylic acid, exhibited a substantial positive correlation with Veillonella parvula on D1. Additionally, Gibberellin A86, another commonly detected urinary metabolite in both D1 and D2 samples, displayed a noteworthy positive correlation with Bacteroides caccae on D1 ( Figure 5C ). The fitting curves of the significant correlations between gut microbiota and blood metabolites, as well as urine metabolites, can be found in Supplementary Figure S6 .

The results suggest an overlap between blood metabolites and urinary metabolites, primarily at D0 and D2. The N6, N6, N6-Trimethyl-L-lysin, which stably exists in the blood metabolites within 48h, is also expressed in D2 urine, and may correlate with specific gut microbiota. Similarly, the KEGG pathway lysine degradation, which is stably present in the blood within 48 hours, is also expressed in D2 urine.

We explored antimicrobial resistance genes in the gut microbiota utilizing the CARD. The Venn diagram in Figure 6A illustrates the detection of 157 ARGs in the gut microbiota of the SA-AKI group and 150 ARGs in the non-AKI group, with 117 genes common to both groups. We identified 65 significantly differentially expressed ARGs based on our screening criteria (adjusted P < 0.05 and |log2FC| > 1). Figure 6B illustrates the heatmap of the number of significantly differentially expressed ARGs. We conducted a comparative analysis of the expression of GO entries with a P < 0.05. The primary biological processes identified included Opsonization, Negative regulation of macrophage-derived foam cell differentiation, and Negative regulation by the host of viral processes ( Figure 6C ). In terms of molecular function, the most significant enrichment and meaningful terms are opsonin binding, suggesting a functional role at the molecular level ( Figure 6D ). However, no significant enrichment was observed in terms related to cellular components. We identified KEGG pathway enriched in sets of differentially expressed ARGs using Metascape. The top KEGG terms included “Valine, leucine and isoleucine degradation,” “Fatty acid metabolism,” “Human papillomavirus infection” and “Neuroactive ligand-receptor interaction” ( Figure 6E ). It merits a particular emphasis that the lysine degradation, a KEGG pathway common to both blood and urinary metabolites, exhibits an association with the citrate cycle (TCA cycle). Supplementary Figure S7 displays an UpSet diagram illustrating the shared KEGG enrichment pathways among blood metabolites, urine metabolites, and antibiotic-resistance genes in the gut microbiota. KEGG pathway analysis shows no overlap between urinary metabolites and gut microbiota ARGs. On D0, blood metabolites and ARGs overlap in Melanogenesis and Vibrio cholerae infection pathways. By days 1 and 2, overlaps include the AMPK and Glucagon signaling pathways, with day 2 adding Valine, leucine, and isoleucine degradation. The interaction among KEGG pathways suggests a potential correlation between blood and urinary metabolites and ARGs in the gut microbiota.