Successful establishment of deep burn models

Deep burns result in damage to the entire epidermal and dermal skin structure, which is highly susceptible to infection and can lead to shock and immune system problems. Mice are in the acute phase on day 1 post-burn and are physiologically hypermetabolised, so we performed rapid rewarming to prevent death of the mice. We observed the burned area compared to normal skin under the microscope (Olympus bx51T) by HE staining and found full thickness necrosis, collagen fibre disorganisation and vacuoles in the epidermis and dermis of the burned area. We also observed vasodilation and cognitive behavioural deficits in the burned group. These phenomena indicate that we have successfully established a mouse model of deep dermal burn and that sampling at 4 and 6 h ensured that samples were obtained in the acute phase.

Statistics of microbiome sequencing results

Intestinal content samples were collected from 16 burned mice (H4 and H6) and 8 normal mice, and DNA was extracted for library construction, which was successful except for one sample in the H4 group. High-throughput 16s rRNA pyrosequencing was then performed on the 23 libraries, yielding a total of 30,281 ASV signature sequences. In terms of family, genus and species, an average of 1,654, 577 and 112 sequences were identified in the H4 group, an average of 1,991, 528 and 97 sequences were identified in the H6 group and an average of 1,088, 264 and 63 sequences were identified in the control group (Table S1 and Figure S1 A and B). These results were used for further analyses. The 23 microbiome datasets were counted at the phylum level, and it was found that the microbial composition of the first sample in the H4 group was significantly different from the other samples, and the number of ASVs was significantly lower than that of the other samples (Figure S1 C and D), which may be caused by insufficient DNA extraction. Therefore, we excluded data from this sample from the control analysis, but included it in the functional analysis and machine learning classification study.

Deep burns lead to dysbiosis of the microbial community in the acute phase

At the phylum level, the major bacterial phyla showed clear trends at 4 and 6 h post-burn. The abundance of the phylum Firmicutes and Actinobacteria declined rapidly at 4 h post-burn, and then further at 6 h (p < 0.05). The opposite trend was observed for Bacteroidetes, Verrucomicrobia and TM7, whose relative abundance continued to increase at both 4 and 6 h (p < 0.05, Fig. 1A). The Firmicutes/Bacteroidetes (F/B) ratio was 1.71 in the control group, while it decreased to 1.05 and 0.95 in the H4 and H6 groups, respectively (Figure S1B). The abundance of another major bacterial phylum, Proteobacteria, showed little fluctuation. At the genus level, the H4 and H6 groups had similar compositions, with no significant differences observed except for Akkermansia (p < 0.05). Compared with the control group, the relative abundances of Oscillospira, Bacteroidaceae_Bacteroides, Akkermansia, Odoribacter, and Mucispirillum were significantly up-regulated after the burn injury, while the abundances of Lactobacillus, Allobaculum and Bifidobacterium were significantly down-regulated (p < 0.05). The abundance of Akkermansia continued to increase after the burn injury, with the H6 group showing a significant increase compared to both the H4 group and the control group; while the Desulfovibrio remained essentially unchanged across the three groups (Fig. 1B). The ASV/OTU Venn diagram clearly illustrates the different numbers of ASVs/OTUs between the control group, the 4 h post-burn group, and the 6 h post-burn group.

Fig. 1

Analysis of the composition and alpha diversity of the intestinal flora of mice in the acute phase of severe burns. (A) Comparison of microbial composition at the phylum level (C: control, H4: 4-hours post-burn; H6:6 h post-burns); (B) Comparison of composition at the genus level; (C) Comparison of α diversity indices between groups ; (D) microbial abundance curves ;E) microbial rarefaction curve

We compared the α-diversity indices of the gut microbiota among the 23 groups. As shown in Fig. 1C, significant differences (p < 0.05) were observed in chao1, Simpson, observed-species, and Shannon indices between the pre-burn and post-burn samples. However, the differences in these indices between the 4-hour and 6-hour post-burn samples were relatively small. The microbial abundance curves suggested that higher number of ASVs were detected in the burn group than in the control group(Fig. 1D). In the rarefaction curve, we observed a significant increase in the diversity of the gut microbiota after the burn injury, with a slight increase in diversity as the acute burn phase progressed. The results reached a stable state when the sequencing depth reached around 30,000 (Fig. 1E).

For β diversity, we performed a PCOA analysis, which revealed a clustering tendency between samples in the distance matrix (Fig. 2A and D). Hierarchical cluster analysis showed clear separability between the control and burn groups (Fig. 2A), indicating significant changes in gut microbiota species richness before and after burn injury. The largest difference was observed between the 6 h post-burn group and the control group, indicating an acute response of the mice to the burn injury. ANOWA analysis showed no significant difference between H4 and H6 (p = 0.143) (Fig. 2B), but both were significantly different from the control group (p < 0.05). We also performed NMDS analysis (Fig. 2C) and found that the differences in microbial community clustering between pre- and post-burn samples were more significant than those observed in PCoA, with a stress value much lower than 0.2, indicating the reliability of the analysis results. These results indicate an increase in species richness and diversity of gut microorganisms during the acute phase of burns, in contrast to other reports that indicate a decrease in diversity.

Fig. 2

Analysis of differences in the gut microbiome between the burn and control groups. (A) Heuristic clustering of 22 mice gut microbiome samples (group C: C1-C8, group H4: H4_2-H4_7; group H6: H6_1-H6_8); (B) Boxplots of microbial abundance in burn and control groups; (C)NMDS analysis of the gut microbiome between the burn and control groups; (D) PCoA analysis of the gut microbiome between the burn and control groups

Characterisation of microbial changes in the acute phase after burns

The LEfSe analyses (Fig. 3A and B) revealed several significantly different bacteria between the burn and control groups. Taxonomically, bacteria from several genera of the phylum Campylobacterota, the phylum Deferribacterales, the phylum Verrucomicrobiota, the order Bacteroidales, and the order Eubacteriales were significantly increased after burn injury, including four genera from group H4: Oscillospira Ruminococcus, Bilophila, Mucispirillum and Helicobacter, and five genera from group H6: Akkermansia, Bacteroides, Anaerotruncus, Odoribacter and Parabacteroides. Multiple groups of bacteria from phylum Actinobacteria, phylum TM7, and class Bacilli are more abundant in normal mice, particularly at the genus level including Lactobacillus, Allobaculum, and Bifidobacterium. Species crossover Venn diagrams also show differences in the sequences detected on the three sets of samples, with the sequences shared between the two burn groups far exceeding those shared between them and the control group (Fig. 3C).

Fig. 3

Analysis and screening of differential microorganisms between burn and control groups. (A) Phylogenetic trees of significantly different species obtained by LEfSe(LDA Effect Size) analysis (LDA ≥ 3.0); (B) Histogram of significantly different species obtained from LEfSe analysis; (C) Venn diagram of average microbial population obtained by sequencing of three groups; (D) Heatmap of the intestinal flora of 22 samples at the genus level (wards.D2 algorithm was used for both species clustering and sample clustering); (E) Heatmap of the intestinal flora of the burn and control groups at the genus level (wards.D2 algorithm was used for both species clustering and sample clustering); (F) Ranking of importance of bacterial genus for classifying three groups (Random Forest algorithm)

The samples and groups were clustered using the 20 most distinct bacteria at the genus level, and the resulting heat maps are shown in Fig. 3D and E. They clearly distinguish burned mice from normal mice and define the three groups. These results are consistent with the results of the LEfSe analysis and can be divided into three cases. The first are those with a significant decrease in abundance after burns, such as Ruminococcus, Lactobacillus, Sutterella, Dehalobacterium, Bifidobacterium, Allobaculum, Adlercreutzia; the second are those with a significant increase in abundance after burns, such as Akkermansia, Parabacteroides, Mucispirillum, Ruminococcus, Alistipes, Bilophila, Bacteroides, Odoribacter, Oscillospira, Helicobacter; and the third are those with a sharp increase at 4 h after burns followed by a recovery starting from 6 h onwards, such as Desulfovibrio and AF12. We used a random forest model to determine the importance of different bacteria in distinguishing between the three groups of samples, and the top bacteria included Bilophila, Parabacteroides, Lactobacillus, Blautia, Odoribacter, Anaerotruncus, and others (Fig. 3F).

Response of microbial functions after burn injury

We extrapolated the functions of differential bacteria to determine which functional adjustments were primarily involved in the changes in gut microbiology after burn injury. We found that the bacteria that changed after burn injury were mainly enriched in several classes of metabolic processes and also involved in functions such as repair, transcription, cell growth and death; in terms of diseases, they were mainly involved in infection, immunity and neurological damage (Fig. 4A). Specifically, lipoic acid metabolism, citrate cycle, seleno-compound metabolism and carbon fixation pathway were enhanced, whereas galactose metabolism, glycerolipid metabolism were weakened in the acute period after burns (Fig. 4B and C). In the 6-hour burn group, amino acid degradation, folate biosynthesis, lipopolysaccharide biosynthesis, vitamin B6 metabolism, and glycosaminoglycan degradation were also enhanced, and the pentose phosphate pathway was weakened (Fig. 4C). In contrast, only lysosome and drug metabolism differed between the H4 and H6 groups (Fig. 4D).

Fig. 4

Function and pathway analysis of differential bacteria in burn and control groups. (A) Metabolic pathways of differential bacterial enrichment between three groups (by number of species); (B) Significantly different metabolic pathways and 95% confidence intervals between the 4-hour post-burn group and the control group; (C) Significantly different metabolic pathways and 95% confidence intervals between the 6-hour post-burn group and the control group; (D) Significantly different metabolic pathways and 95% confidence intervals between the 4-hour post-burn group and the 6-hour post-burn group

Machine learning models accurately discriminated burn phenotypes

In the model building process, the top twenty biomarkers were first obtained, and then classification was performed. Both XGBoost and Random Forest achieved 100% accuracy in classifying the control and burn groups (Fig. 5A-D). SVM achieved 86.7% accuracy in classifying the 4 and 6 H groups (Fig. 5E-H). These results demonstrate that it is possible to predict the symptoms and severity of burn injury in mice by using specific microbial abundances as feature values. The slightly lower prediction accuracy for Burn 4 H and Burn 6 H is due to the fact that both time points are in the acute phase of burn when changes in the microbial community are minimal. However, these trends can still be observed.

Fig. 5

Results of machine learning classification of three groups of samples. (A) Confusion matrix for the classification of burn and control groups using the XGBoost algorithm; (B) Confusion matrix for the classification of burn and control groups using the Randomforest algorithm; (C) Confusion matrix for the classification of burn and control groups using the Gaussian Naive Bayes algorithm; (D) ROC curves for classification of burn and control groups; (E) Confusion matrix for the classification of 4-hour post-burn group and the 6-hour post-burn group using the Randomforest algorithm; (F) Confusion matrix for the classification of 4-hour post-burn group and the 6-hour post-burn group using the SVM algorithm; (G) Confusion matrix for the classification of 4-hour post-burn group and the 6-hour post-burn group using the NaiveBayes algorithm; (H) ROC curves for classification of two burn groups