Abstract

   Acute pancreatitis (AP) is an inflammatory disease of the pancreas.
   Despite of a steadily increasing in morbidity and mortality, there is
   still no effective therapy. Gut microbial dysbiosis and its
   derived-metabolites disorder have been shown to play an important role
   in the development of AP, however, little is known regarding the
   crosstalk between gut microbiota and metabolites. In this study, we
   assessed the alterations in gut microbiota and metabolites by
   constructing three AP mouse models by means of metagenomic and
   metabolomic sequencing, and further clarified their relationship by
   correlation analysis. The results revealed that each model exhibited
   unique flora and metabolite profiles. KEGG analysis showed that the
   differential flora and metabolite-enriched pathway functions were
   correlated with lipid metabolism and amino acid metabolism. Moreover,
   two core differential bacterial species on Burkholderiales bacterium
   YL45 and Bifidobacterium pseudolongum along with eleven differential
   metabolites appeared to exert certain effects during the course of AP.
   In conclusion, further exploration of the crosstalk between microbiota
   and derived metabolites may provide novel insights and strategies into
   the diagnosis and treatment of AP.

   Keywords: Acute pancreatitis, Metagenomics, Metabolomic, Gut microbiota

   Subject terms: Microbiology, Pancreatic disease

Introduction

   Acute pancreatitis (AP) is a common gastrointestinal disease that
   contributes to hospitalization in emergency^[38]1. In recent years, the
   role of gut microbial dysbiosis in the pathogenesis of AP has received
   extensive attention^[39]2. In the occurrence of AP, excessive
   inflammatory cytokines are released into the bloodstream, and then a
   series of physiological activities are triggered including gut barrier
   damage, dysbiosis of gut microbiome, and bacterial translocation^[40]3.
   In turn, the bacteria entering the blood through the intestinal wall
   may influence several organs. There is considerable evidence that
   bacterial diversity reveals a significant decrease in AP, while the
   abundance and diversity of gut microbiota are closely related to the
   gut barrier^[41]4. Also, with the progression of AP, the imbalance of
   gut flora has a significant impact on the intestinal barrier function
   to exacerbate disease^[42]5,[43]6. In addition, evidence has
   demonstrated that the severity of AP presents a certain relationship
   with the composition of gut microbiome^[44]7. When compared with mild
   acute pancreatitis (MAP) and moderately severe acute pancreatitis
   (MSAP) patients, severe acute pancreatitis (SAP) patients have shown a
   lower abundance of beneficial bacteria such as Bacteroides,
   Alloprevotella, Blautia, and Gemella in fecal samples^[45]8. With the
   continuous improvement of advanced sequencing methods such as
   next-generation sequencing and metagenomics, a more in-depth
   understanding of the key role of gut microbiota in human health and
   disease has been observed^[46]9. As of now, metagenomics has been
   employed to identify the etiological factors, investigate taxonomic
   classifications, and determine the underlying mechanisms of AP.

   Gut microbiota is involved in the progression of AP by repairing the
   gut barrier as a result of regulating host metabolism and immune
   response^[47]10, while gut microbiota-derived metabolites are crucial
   segments that participate in intestinal-pancreatic axis
   interactions^[48]11. It is widely known that gut microbiota elicits
   superior metabolic activities with the conversion from host and dietary
   sourced components to a variety of metabolites, some of which take
   effect not only in the gut but also in other targeted organs far from
   the intestine through the circulation, resulting in great influence on
   physiological and pathological processes^[49]12. To date, the key role
   of gut microbiome in AP has been extensively demonstrated, and relative
   effects of short-chain fatty acids (SCFAs) and bile acid metabolism
   have also been illustrated^[50]13,[51]14. However, there is limited
   knowledge of pertinent literature about non-classical metabolites
   generated from gut flora. Also, the links between microbiota and
   metabolites remain complicated and unclear, which needs further
   exploration. With the critical role of the gut microbiome and its
   metabolites becomes apparent, identifying new microbial-metabolite
   combinations and their relationship to AP pathogenesis should be noted.

   In this study, we explored the species composition and characteristics
   of the structure of intestinal microbiota in three AP mouse models.
   Through metagenomics and untargeted metabolomics analysis, we
   characterized the bacterial flora of AP and revealed the correlation
   between gut microbiota and metabolites, which could provide insights
   into elucidating the underlying mechanisms of gut microbiota and
   metabolites on the course of AP.

Results

Establishment of three AP mouse models

   To assess whether the models were successfully constructed, pancreatic
   histopathology, cytokines, and serum enzyme levels (lipase and amylase)
   were evaluated. In the Caerulein group, widening of the lobular septum
   and cellular gap, edema, vacuolization and necrosis of pancreatic
   acinar cells, and infiltration of inflammatory cells were observed
   microscopically in the pancreas. In the Caerulein + LPS group, the
   pathologic features of the pancreas were similar to those in the
   Caerulein group and the infiltration of inflammatory cells was more
   obvious than that in the Caerulein group. As for the L-arginine group,
   more severe pathological features were found in the microscopy, that
   is, diffuse necrosis of acinar cells, marked edema of the lobular
   septum, and infiltration of inflammatory cells (Fig. [52]1A). The
   histopathologic scores of the pancreas in the three groups of AP mice
   were significantly higher than the Control group (Fig. [53]1B) as well
   as showed a significant difference of serum amylase, and lipase levels
   (Fig. [54]1C,D). Additionally, the number of Ly6G + neutrophils and
   CD68 macrophages in the pancreatic tissues of AP revealed a highly
   significant increase compared to the Control group, along with
   elevation of the serum inflammatory cytokines IL-1β, IL-6, and TNF-α
   (Fig. [55]1E–I). The above results indicated the presence of pancreatic
   injury and systemic inflammatory response in AP mice, suggesting that
   three AP mouse models were successfully established.

Fig. 1.

   [56]Fig. 1
   [57]Open in a new tab

   Establishment of three AP mouse models. (A) HE staining of
   histopathological images of pancreatic tissues among the four groups.
   (B) Histopathological scores of pancreatic tissues in different groups.
   (C,D) Serum amylase and lipase levels. (E,F) Immunohistochemical
   pictures of pancreatic tissues among the four groups with Ly6G and
   CD68. (G–I) Serum levels of inflammatory cytokines IL-1β, IL-6, and
   TNF-α in different groups. N = 6 mice. Data are presented as mean ± SD.
   AP model groups were compared to the Control group, *P < 0.05;
   **P < 0.01; ***P < 0.001; ****P < 0.0001.

Alterations in intestinal inflammation and barrier function in AP mouse
models

   Intestinal barrier damage is often accompanied during the onset of AP,
   and to understand the effects on the gut, HE staining was used to
   assess the pathology of the colonic mucosa in four groups, which showed
   normal structure in Control group, with intact and regular intestinal
   villus structure, and no edema in the interstitium. In the Caerulein,
   Caerulein + LPS, and L-arginine groups, it demonstrated a relatively
   intact intestinal epithelium and regular villi arrangement, but the
   intestinal muscularis was thinned, and the submucosa was dilated and
   accompanied by lymphoid follicular hyperplasia (Fig. [58]2A). RT-qPCR
   was employed to identify the relative mRNA expression of IL-1β, IL-6,
   and TNF-α in colon tissues. Compared with the Con group, the mRNA
   levels of pro-inflammatory cytokines were significantly increased in
   the Caerulein group and Caerulein + LPS group, while there were no
   significant changes in the L-arginine group (Fig. [59]2B–D). The
   integrity of intestinal villi is closely related to intestinal
   inflammation and barrier function. Therefore, the protein expression
   levels of ZO-1, Occludin, and Claudin 1 in the colon of four groups
   were examined by Western Blot, and the results showed that compared
   with the Control group, the protein levels of ZO-1 and Occludin were
   significantly reduced in the three AP groups, and concurrently the
   protein level of Claudin 1 was also demonstrated a highly significant
   reduction in the Caerulein + LPS group (Fig. [60]2E,F). Based on these
   findings, the intestinal mucosal barrier function of mice was damaged
   to different degrees during the development of AP.

Fig. 2.

   [61]Fig. 2
   [62]Open in a new tab

   Comparison of intestinal barrier function in three AP mouse models. (A)
   HE staining of pathological images of colon tissues among the four
   groups. (B–D) Relative mRNA levels of IL-1β, IL-6, and TNF-α in colon
   tissues. (E, F) Protein expression levels of ZO-1, Occludin, and
   Claudin 1 in colon tissues. The uncropped blots are shown in the
   supplementary information file. N = 6 mice. Data are presented as
   mean ± SD. Statistically significant differences were marked with:
   *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 when compared with
   Control group.

Analysis of gut microbiome diversity

   Alpha diversity specifically refers to the diversity and abundance of
   species in a community and is mostly used to assess the differences in
   species diversity within a colony. Results demonstrated (Fig. [63]3A–C)
   that in comparison with the Control group, the Chao 1 index of the
   L-arginine group decreased significantly, the Shannon index of the
   Caerulein and L-arginine groups revealed a highly significant
   reduction, and along with a marked increase of Simpson index in the
   Caerulein + LPS group (P < 0.05). Beta diversity reflects the
   differences in the composition of the species diversity in different
   communities, which are always measured according to PCoA to assess the
   sample distance between two groups to reflect the structure of the gut
   microbiota. It provided an overview of the gut microbiome that
   Caerulein, Caerulein + LPS, and L-arginine groups could be clearly
   distinguished from the Control group (Fig. [64]3D–F), indicating that
   all three AP models differed from the Control group (P < 0.05). ANOSIM
   analysis is a statistical method to analyze the similarity between
   groups of high-dimensional data. The results of this study are shown in
   Fig. [65]3G–I. Separate comparisons between three AP models and Control
   group were performed with the R values of 0.997, 0.381, and 0.407,
   accompanied by the P values of 0.003, 0.001, and 0.003, indicating that
   all three AP models displayed a difference with the Control group.

Fig. 3.

   [66]Fig. 3
   [67]Open in a new tab

   Diversity analysis of gut microbiota in three AP models. (A) Chao 1
   index. (B) Shannon index. (C) Simpson index. (D–F) PCoA plots showed
   statistically significant differences between the Control and AP groups
   respectively. (G–I) ANOSIM plots showed statistically significant
   differences between the Control and AP groups. N = 6 mice.

Characterization of the species composition of the gut microbiota in AP and
enrichment analysis of related functional genes

   Characterizing the abundance of gut microbiota at different annotation
   levels in a relative quantitative perspective, when compared with the
   Control group, the results showed the abundance of the top ten phyla in
   different groups, as shown in Fig. [68]4A. Analysis of the composition
   of microbiomes (ANCOM) revealed that Proteobacteria was upregulated in
   the three AP models along with an increase in Bacteroidetes in the
   Caerulein group and L-arginine group. As for the Caerulein group,
   Firmicutes demonstrated a significant downregulation, while the
   contrary tendency appeared in L-arginine group (Fig. S1, Table [69]S1).
   Then, the composition of the top ten species was analyzed, and the
   results are shown in Fig. [70]4B as described by differences in the
   species composition among the four groups. Furthermore, the histograms
   of the distribution of linear discriminant analysis (LDA) values
   obtained from the LEfSe analysis (Fig. [71]4C–E, LDA > 3.0 as the
   screening criterion) also revealed differences in microbial abundance
   between the Control group and the AP models. Collectively, these
   results suggested that species composition at the phylum and species
   levels could be influenced in AP models. KEGG analysis was utilized to
   explore the main signaling pathways involved of the differential
   functional genes, and after that found metabolism (i.e., amino acids,
   carbohydrates, and lipids) and information processing pathways were
   involved (Fig. [72]4F–H)^[73]15–[74]17.

Fig. 4.

   [75]Fig. 4
   [76]Open in a new tab

   Compositional structure of gut flora and KEGG pathway enrichment
   analysis. (A,B) Relative abundance of the top ten gut microbiota at the
   phylum and species levels. (C–E) LefSe analysis showed predominant gut
   microbiota. (F–H) KEGG enrichment pathways in AP mouse models in
   comparison with the Control group. N = 6 mice.

Multivariate statistical analysis of serum metabolites in AP

   To analyze the differences in metabolic profiles between groups,
   multivariate statistical analysis was used to reduce the dimensionality
   of the complex data. PCA analysis showed that the samples of the
   Caerulein group, the Caerulein + LPS group, and the L-arginine group
   could be clearly differentiated from those of the Control group (Fig.
   S2A-C), which suggested that there existed enlarged differences between
   Control and AP groups and a large number of metabolites were
   differentially expressed with an indirect description of successful
   model construction in this study. Then, to further screen the
   differential metabolites between the Control group and the AP groups,
   the results were analyzed with the aid of OPLS-DA shown in Fig. S2D-F,
   which showed a clear separation between the Control group and the AP
   groups, likewise indicating that there were significant differences of
   the metabolites in the AP groups compared with the Control group.
   Subsequently, OPLS-DA model R2X (cum), R2Y (cum), and Q2 (cum) scores
   were further calculated with the result of 0.347, 0.994, and 0.773
   within Caerulein group in contrast to Control group, in the meanwhile,
   it showed a calculation of 0.437, 0.99 and 0.917 in Caerulein + LPS
   group, followed by 0.336, 0.967 and 0.381 in L-arginine group. Also,
   the accuracy of the OPLS-DA was assessed utilizing the 200 response
   permutation testing (Fig. S2G–I).

Analysis of differential metabolites

   In this study, a total of 512 metabolites were identified and
   classified. Excluding the classification of others, the top 5
   substances were lipids and lipid-like molecules (192, 37.5%), organic
   acids and derivatives (95, 18.6%), organoheterocyclic compounds (65,
   12.7%), organic oxygen compounds (43, 8.4%), and benzenoids (19, 3.7%)
   (Fig. [77]5A). Volcano plots were used to show the trend of the
   expression of all detectable substances in the Control group versus the
   AP model groups, with VIP > 1.0 and P < 0.05 as the differential
   metabolite screening thresholds, showing a large number of differential
   metabolites presented in all three AP model groups compared with the
   Control group (Fig. [78]5B–D). Then, pathway enrichment analysis of
   potential biomarkers and screening of pathways with impact values
   greater than 0 were performed using the MetaboAnalyst system. As shown
   in Fig. [79]5E–G, the primary pathways included arginine and proline
   metabolism, cysteine and methionine metabolism, linoleic acid
   metabolism, valine, leucine, and isoleucine biosynthesis, glycine,
   serine, and threonine metabolism, and histidine metabolism, tryptophan
   metabolism, and biosynthesis of unsaturated fatty acids. All of these
   pathways played an important role in a variety of physiological
   functions, suggesting that metabolites were involved in the occurrence
   of AP by affecting normal physiological functions.

Fig. 5.

   [80]Fig. 5
   [81]Open in a new tab

   Differential metabolite analysis and metabolic pathway analysis. (A)
   Pie chart showed the categories of the different metabolites. (B–D)
   Volcano plots showed differential metabolites and VIP values in the
   Control and AP groups. (E–G) Bubble plots of significantly differential
   metabolite pathway analysis in the Control and AP groups. N = 6 mice.

Microbiota-metabolite correlation analysis

   To clarify the relationship between metabolites and gut microbiota,
   spearman's correlation analysis was performed on the differential flora
   (LDA > 3.0 as the screening criterion) and metabolites in the AP model
   groups, and the absolute value of the correlation coefficient > 0.8 was
   further used as the screening criterion to construct the intestinal
   microflora-metabolite correlation network. The results showed
   (Fig. [82]6A–C) that in the Caerulein group, 15 differential strains
   such as Bacteroides thetaiotaomicron were identified as the core
   strains, and 19 differential metabolites such as MethylIndole-3-acetate
   were determined as the core metabolites. As for Caerulein + LPS group,
   6 differential bacteria including Alistipes communis, and 17
   differential metabolites such as 3-Methylglutarylcarnitine were core
   strains and metabolites respectively. Also in the L-arginine group, 10
   differential flora including Alistipes onderdonkii, and 18 differential
   metabolites such as 2-Hydroxystearic acid were recognized as core
   intestinal bacteria and metabolites. Then, the differential strains of
   the three AP model groups (LDA > 2.0 was used as the screening
   criterion) were taken as intersections, and the Venn plots
   (Fig. [83]6D) showed that Burkholderiales bacterium YL45 and
   Bifidobacterium pseudolongum differed significantly among the three
   model groups, and presented an increasing tendency in all AP mouse
   models. Similarly, a total of 16 metabolites were significantly
   different in the three models (Fig. [84]6E), among which 14 metabolites
   changed in the same direction (including 7 up-regulated and 7
   down-regulated), and we then carried out a spearman correlation
   analysis of the two differential strains with the 14 differential
   metabolites, results showed that the two strains were correlated with
   11 differential metabolites (Fig. [85]6F).

Fig. 6.

   [86]Fig. 6
   [87]Open in a new tab

   Microbiota-metabolite interactions. (A–C) Network diagram of a
   differential microbiota-metabolite correlation between Control and AP
   groups. (D–E) Venn plots demonstrated the intersection of the
   differential strains and metabolites in three AP mouse models. (F)
   Heatmap of the correlation between common differential strains and
   metabolites. N = 6 mice. Significant level was marked with *P < 0.05;
   **P < 0.01; ***P < 0.001; ****P < 0.0001.

Discussion

   AP is characterized as an inflammatory response caused by pancreatic
   enzyme activation with multiple etiologies, which leads to
   autodigestion of pancreatic tissues and then secondary edema,
   hemorrhage, and necrosis. Currently, the common modeling methods of AP
   include administration of Caerulein, Caerulein + LPS, and L-arginine.
   In previous studies, researchers focused on convenience for operation,
   reproducibility, and clinical relevance of these 3 models, in order to
   provide guidance for choosing the appropriate AP model^[88]18–[89]20.
   However, to date, the explorations remain single omics or single models
   with the difficulty to completely summarize the clinical phenotype of
   AP because of the limitation of animal models. Thus, further studies
   need to concentrate on multi-omics and models to provide a perfect
   evidence chain for clinical research.

   We constructed three AP mouse models in this study and found that all
   of them showed pancreatic and intestinal barrier injury. In the
   Caerulein and Caerulein + LPS group, we found consistent alterations in
   H&E phenotype and other biochemical markers, however, it was
   interesting that the L-arginine group presented inconsistent results.
   In the previous studies, Zhang et al. found that the amylase and lipase
   of L-arginine-induced AP model were elevated at 6 h, reached the peak
   at 12 h, and then gradually decreased^[90]21. Also, Hegyi et al. found
   that the IL-1β levels increased significantly at 12 h after L-arginine
   injection, peaked at 24 h, and decreased thereafter^[91]22. In this
   study, the time point of observation was considered as 72 h, therefore
   the relevant indexes declined in consistency with previous results.
   However, other literature has also shown that pancreatic injury
   (including amylase and lipase levels) is the severest in the
   L-arginine-induced AP model at 72 h^[92]23. This inconsistency remains
   controversial which needs to collect relevant samples at different time
   points of the model in subsequent experiments to determine the exact
   trend of the changes. As for the intestinal barrier injury, it was
   manifested in pathological damage to the intestinal mucosa and
   down-regulation of tight junction protein levels. Studies have shown
   that intestinal barrier damage is one of the major complications of AP,
   as revealed in a meta-analysis study, 59% of AP patients have found gut
   barrier dysfunction^[93]6. In addition, another study showed that
   bacterial DNA representing gut microbiota could be detected in the
   blood of 68.8% of patients with AP, suggesting that gut microbiota can
   be translocated to the bloodstream through the damaged gut barrier in
   AP^[94]24. Also, the translocation of gut flora can further lead to
   pancreatic tissue necrosis and infection, thus triggering SIRS and
   MODS^[95]25, which is closely related to the high mortality of SAP
   patients.

   In this study, we observed the alteration of gut microbiome in three AP
   mouse models and found good general agreement with previous clinical
   and animal studies in the variation tendency of Firmicutes,
   Bacteroidetes, and Proteobacteria at the phylum level, which reflected
   the reliability and reproducibility of metagenomics sequencing as well
   as the potency of inference to clinical research^[96]26–[97]28. Also,
   an elevation of the abundance of Burkholderiales bacterium YL45 and
   Bifidobacterium pseudolongum was captured in three AP models through
   the comparison at the species level, indicating that they may exert
   crucial role in the progression of AP, while the specific actions need
   to be further explored. As an unclassified member of phylum
   Proteobacteria and genus Burkholderiales, Burkholderiales bacterium
   YL45 has demonstrated persistent existence and occupied a dominant
   position in active inflammatory bowel disease^[98]29. Concurrently,
   another study found its effect on disease deterioration by ileum
   colonization, suggesting the involvement in inflammatory
   diseases^[99]30,[100]31. Bifidobacterium pseudolongum belongs to the
   genus Bifidobacterium, which presents beneficial efficiency on the
   elimination of pathogens and immune regulation^[101]32. As shown in
   non-alcoholic fatty liver disease-associated hepatocellular carcinoma,
   Bifidobacterium pseudolongum could remodel the composition of gut
   microbiome and improve the gut barrier function to prevent occurrence
   of disease^[102]32. Moreover, gut barrier damage was detected to be
   alleviated in mice with colitis by modulating PPARγ/STAT3 pathway and
   gut microbiome composition, restoring antioxidant enzymes’ activities,
   along with suppressing the release of proinflammatory
   cytokines^[103]33. However, we observed a high abundance of
   Bifidobacterium pseudolongum in three AP models in the current study,
   indicating its side effects on AP. Likewise, it was found that the
   abundance of Bifidobacterium pseudolongum was increased with a positive
   correlation to cytokines in influenza^[104]34. In a comprehensive
   consideration, this surprising result might be attributed to the
   heterogeneity of reaction towards diseases, that is specifically
   microbial-mediated protection of the host.

   As a result of metabolomics analysis, a total of 14 metabolites existed
   differences in three AP models and manifested a consistent variation
   tendency. Docosahexaenoic acid (DHA) is an omega-3 polyunsaturated
   fatty acid that has been widely explored in the prevention and
   treatment of many diseases due to its antioxidant and anti-inflammatory
   functions. Besides, relevant animal and cellular studies have found a
   positive effect of DHA in AP at present^[105]35, however, this effect
   is closely related to the concentration with relatively high
   concentrations of DHA inducing Ca^2+-mediated activation of protein
   kinase C (PKC) isoforms (PKC-α, PKC-δ, PKC-ε, and PKC-ζ) and zymogens
   in pancreatic acinar cells, which promotes the progression of
   AP^[106]36. In this study, AP-related metabolites were mainly involved
   in glycerophospholipid metabolism pathway (PC (22:5 (7Z, 10Z, 13Z, 16Z,
   19Z)/20:1 (11Z)), PC (22:2 (13Z,16Z)/15:0), LysoPC (20:0/0:0), PC (18:2
   (9Z,12Z)/14:0)) with a close relationship to the development of
   diseases. Phosphatidylcholine (PC) is an important biomolecule whose
   molecular structure contains a polar phosphatidylcholine head and two
   nonpolar fatty acid hydrocarbon chains. As a category of the core
   phosphoglycerolipid metabolites, they have been identified as a key
   player in the composition of mammalian cell membranes and are closely
   associated with a variety of physiological processes^[107]37,[108]38.
   When it comes to Lysophosphatidylcholine (LPC), a major component of
   oxidatively damaged low-density lipoproteins, it shows strong
   bactericidal and anti-inflammatory effects. In infectious diseases such
   as sepsis, peritonitis, and pneumonia, LPC inhibits the inflammatory
   response by suppressing the production of inflammatory mediators and
   increasing the levels of anti-inflammatory factors^[109]39,[110]40.
   Additionally, Linoelaidic acid (LA) is classified as a major component
   of LDL and an essential fatty acid, which can be oxidized by
   circulating endogenous enzymes and reactive oxygen. It has been
   reported that levels of LA and arachidonic acid were significantly
   elevated in AP patients^[111]41, and similarly, LA produced from
   lipolysis was observed to be lipotoxic in the SAP mouse model, which
   could contribute to renal tubular injury^[112]42. The above
   differential metabolites in AP have been widely explored and are
   expected to be applied in the diagnosis and treatment of AP, whereas
   their specific regulatory mechanisms and subsequent clinical
   applications remain to be further investigated.

   We performed function analysis based on the gut microbiome-related
   differential genes between the Control group and the three AP groups.
   It was surprising to find a highly consistent result with that of serum
   metabolomics, to be specific, it was mainly associated with several
   signaling pathways, such as metabolism (lipid metabolism, amino acid
   metabolism) and information processing. Disorders of amino acid
   metabolism are involved in a variety of disease processes including AP,
   and several recent studies have shown that targeted amino acid
   supplementation contributes to the remission of AP, in which the
   interaction mechanism involves anti-inflammatory, antioxidant,
   etc^[113]43,[114]44. Amino acid metabolism may be an important
   therapeutic target of AP. Specifically, if we found a certain amino
   acid was decreased/increased in AP, whatever human or mice samples, the
   corresponding supplementation or decrease of this amino acid may
   relieve AP. In the present study, we identified multiple alterations in
   amino acid metabolism. Valine, leucine, and isoleucine are the main
   components of branched-chain amino acids (BCAAs) with an elevated
   expression in necrotizing AP models^[115]45, and their production of
   succinyl coenzyme A and acetyl coenzyme A can enter the tricarboxylic
   acid cycle or through the process of gluconeogenesis to generate energy
   to meet the high catabolic state of AP^[116]46. However, the specific
   actions and in-depth mechanisms targeting the amino acids need to be
   further explored. Evidence has existed to support the role of enhancing
   the expression of tight junction proteins, and mucosal immune function
   of BCAAs, together with being potential biomarkers for a variety of
   diseases. In the present study, in contrast, both metagenomic and
   metabolomic analyses revealed that AP was linked with the synthesis of
   BCAAs, prompting the possibility of the utilization as biomarkers and
   therapeutic targets for AP^[117]47. Regarded as significant parts of
   protein synthesis, methionine and cysteine are engaged in the
   regulation of sulfur metabolism as well as glutathione
   synthesis^[118]48. During the development of AP, methionine and
   cysteine metabolism is dysregulated owing to dysregulation of
   trans-sulfur regulation mediated by nitration of
   cystathionine-β-synthase, resulting in a rapid depletion along with a
   significant reduction in glutathione levels^[119]49, whereas
   glutathione depletion has determined to be an accordant feature of the
   progression from MAP to SAP which can be attributed to aggravating
   oxidative stress and impaired upstream and downstream amino acids
   biosynthesis^[120]50. Arginine metabolism promotes NO production in the
   early stages of acute inflammation, and the combined impairment of
   arginine and NO metabolism has the chance to lead to the disruption of
   intestinal barrier function and progression of inflammation^[121]51.
   Simultaneously, a succession of catabolic processes take place
   including catabolizing into ornithine and urea under the action of
   arginase, and then ornithine is further decarboxylated to polyamines
   and proline^[122]52. Proline has been discovered to improve human
   immune defense and resistance to foreign antigens and pathogens through
   gut-associated lymphoid tissues, while polyamines can regulate the
   integrity of the intestinal epithelial barrier by controlling the
   expression of tight junction proteins and mucosal adaptive immunity, as
   well as suppressing the level of inflammation and oxidative
   stress^[123]53. All above series of cascade reactions indicate that the
   regulation of arginine-proline metabolism has the potential to be a
   significant target for future prevention of intestinal barrier
   dysfunction and thus showing positive effects on the treatment of AP.

   However, this study has some limitations, including small sample size
   and lack of clinical validation, thus further studies with enlarged
   sample size and clinical investigation should be carried out to confirm
   our findings. Moreover, specific animal and cell experiments are
   warranted to validate the effect of targeted flora and metabolites in
   AP. Besides, this study focused on the acute phase of AP, whereas for
   better analysis, future studies should conduct longitudinal
   observations and investigate potential dynamic changes in the gut
   microbiota throughout the overall course of AP.

   Overall, this study aimed to evaluate the levels of gut microbiome and
   serum metabolites in different AP mouse models to determine the
   function of gut flora-derived metabolites. Interestingly, we found that
   the composition of gut microbiota and metabolites in the three AP
   models altered significantly in comparison to the Control group, and
   the trend of change showed unique performance among different models.
   Besides, analysis of different strains and metabolite functions mainly
   involved the abnormal pathways of lipid metabolism and amino acid
   metabolism. Moreover, Burkholderiales bacterium YL45 and
   Bifidobacterium pseudolongum as the main intestinal bacteria appeared
   to be a reasonable choice to improve the inflammatory responses of AP.
   Also as expected, targeting gut microbiota and derived metabolites may
   be a promising avenue for novel treatment development.

Material and methods

Animal experiments and ethical approval

   Male C57BL/6 J mice (aged 7 weeks and weighing 20–25 g) were provided
   by the Department of Zoology, Central South University. The mice were
   housed under specific-pathogen-free (SPF) conditions with free ad
   libitum access to food and water and maintained in a regular 12 h
   light–dark cycle. A total of three AP mouse models were employed in the
   experiments, in which the Caerulein group was induced by hourly
   intraperitoneal (i.p.) injection of caerulein (200 μg/kg, once an hour
   for a total of 10 times) (MedChemExpress, USA), and the Caerulein + LPS
   group was administered by hourly i.p. injection of caerulein
   (200 μg/kg, once an hour for a total of 10 times) (MedChemExpress, USA)
   combined with a dose of 5 mg/kg lipopolysaccharide (LPS)
   (Sigma-Aldrich, USA), mice in the L-arginine group received 2 times
   i.p. injection of L-arginine (4 g/kg, once an hour, twice in total)
   (Sigma-Aldrich, USA). Similar saline injections were applied in the
   Control group. Mice in the Caerulein group, Caerulein + LPS group, and
   Control group were sacrificed 12 h after the first caerulein or saline
   injection, while those in the L-arginine group were executed at 72 h.
   After that, pancreatic tissues were isolated, also blood and feces were
   collected for further analysis. All animal experimental protocols were
   approved by the Ethics Committee of Central South University (No.
   FORM-IACUC-01–01) before the experiments. All the experimental
   operations were performed in accordance with relevant regulations and
   ARRIVE guidelines.

Measurement of serum enzymes and pro-inflammatory cytokines

   Serum was obtained by centrifuging the blood samples at 3000 rpm for
   20 min at a temperature of 4 °C. Serum amylase, lipase, and IL-1β,
   IL-6, and TNF-α levels were detected using commercial kits purchased
   from ZCIBIO Technology Co., Ltd. (Shanghai, China) as detailed
   procedures followed the manufacturer's protocol.

Histological analysis and immunohistochemistry (IHC)

   To carry out the histologic evaluation of hematoxylin–eosin staining,
   the dissected pancreatic tissues were fixed, dehydrated, and then
   embedded in paraffin. Thereafter, the embedded paraffin sections were
   cut to a thickness of 3 μm and followed by being stained with
   hematoxylin and eosin (H&E). For IHC assays, the paraffin sections were
   dewaxed, endogenous enzymes inactivated and antigens thermally
   repaired. The sections were then blocked and stained with CD68 or Ly6G
   antibodies, followed by staining with the appropriate secondary
   antibodies and hematoxylin. Subsequently, the morphology and structure
   of the stained sections were observed using a Zeiss microscope.
   Pancreatic histologic scores were calculated in the aspects of edema,
   inflammation, hemorrhage, and necrosis as described previously.

RNA isolation and quantitative real-time PCR

   According to the instructions of the manufacturer, RNA was meticulously
   extracted from the colon tissues using TRIzol (Vazyme). Then, we
   employed the cDNA Synthesis Kit from Vazyme (RC323-01) to conduct the
   reverse transcription and then performed real-time PCR analysis with
   ChamQ Universal SYBR qPCR Master Mix, also provided by Vazyme
   (Q711-02/03). To ensure accurate calibration, β-actin was selected as
   the internal control to calibrate the original mRNA expression levels.
   All sequences of primer used in this research are listed in
   Table [124]S2.

Western blot analysis

   The colon was cryogenically ground in liquid nitrogen and lysed in RIPA
   buffer supplemented with protease and phosphatase inhibitors. The
   protein concentration in the colon was measured using a BCA protein kit
   (Beyotime, China). After heating at 100 °C for 10 min, equal amounts of
   protein were separated by SDS-PAGE and then transferred onto a PVDF
   membrane. The PVDF membrane was incubated in 5% bovine serum albumin
   for 2 h at room temperature to seal the non-specifically binding
   proteins, incubated overnight at 4 °C with primary antibodies, and
   finally with secondary antibodies incubated for 1 h at room
   temperature. All antibodies used were purchased from Proteintech Group,
   China. The protein bands were visualized by electrochemiluminescence
   (ECL) assay kit (Beyotime, China) and then analyzed the density using
   Image Lab. Relative protein levels were standardized to the β-actin.

Metagenomics

   Fresh fecal samples were collected from mice and immediately frozen in
   liquid nitrogen. DNA extraction was performed using the CommaXP® Fecal
   Genomic DNA Extraction Kit, and sequencing was conducted using the
   MGISEQ-2000 sequencer (MGI Tech Co., Ltd.). To ensure the accuracy and
   reliability of the data, KneadData software (v0.10.0,
   [125]https://huttenhower.sph.harvard.edu/kneaddata/) was used for
   quality control (QC). Briefly, the software employed fastqc for QC,
   then trimmomatic for filtering, bowtie2 to compare the host genome to
   remove the host, and finally fastqc to perform QC again. The entire
   valid sequences of all samples were annotated and categorized by
   Kraken2 (v2.0.8-beta). Information on species abundance at each level
   was obtained by comparing CleanData with species sequences in the
   database. The alpha diversity indices were calculated to analyze the
   complexity of the species diversity, including the Chao1 index, Shannon
   index, and Simpson index by using the vegan package from R software
   (version 2.15.3). Then, analysis of similarities (ANOISM) was analyzed
   using the anoism function of the vegan package. Distances were
   calculated using the vegan package and then analyzed by PCoA using the
   cmdscale. Analysis of composition of microbiomes (ANCOM) was employed
   to identify the relevant bacteria in each model using the ANCOMBC
   package. Also, the different groups were assessed regarding the
   variation in the abundance of microbiota by Linear discriminant
   analysis effect size (LEfSe) and then visualized with histogram plots.
   Analysis of metabolic pathways presented in the microbiome was
   performed using HUMAnN2.

Untargeted metabolomics

   50 μL of serum sample was pipetted into an EP tube, mixed with 150 μL
   of extraction solution (methanol/acetonitrile = 1:1(v/v) containing
   isotope-labeled internal standard mixture) to separate metabolites.
   Mass spectrometry analysis was performed using a Vanquish (Thermo
   Fisher Scientific) ultra-performance liquid chromatograph in positive
   and negative mode. Quality control (QC) samples were used to ensure
   stability and reproducibility. Raw data were converted to mzXML format
   by ProteoWizard software (v3.0.24054-2352758 (automated build),
   [126]https://hub.docker.com/r/proteowizard/pwiz-skyline-i-agree-to-the-
   vendor-licenses/tags?page=1&page_size=&name=&ordering=/), processed for
   peak identification, peak extraction, peak alignment, and integration,
   and then matched with BiotreeDB (V2.1) self-constructed secondary mass
   spectrometry database for substance annotation. The final dataset
   containing peak number, sample name, and normalized peak area
   information was imported into the SIMCA16.0.2 software package
   (Sartorius Stedim Data Analytics AB, Umea, Sweden) for multivariate
   analysis. The R package MetaboAnalystR was used for principal component
   analysis (PCA), and orthogonal partial least squares discriminant
   analysis (OPLS-DA). To check the robustness and predictive ability of
   the OPLS-DA model, 200 permutations were further conducted. Variable
   Importance in the Projection (VIP) of the first principal component in
   OPLS-DA analysis was also acquired. Metabolites with VIP > 1 and
   P < 0.05 were considered significantly altered metabolites.
   MetaboAnalyst and KEGG were used for further pathway analysis.

Statistical analysis

   All data were statistically analyzed using GraphPad Prism 9.0 software.
   As for two independent samples, the Student's t test was used for
   analysis for those who conformed to normal distribution, otherwise
   Mann–Whitney U test was used. In the comparison of multiple groups,
   One-Way ANOVA analysis was applied for those that conformed to normal
   distribution and homogeneity of variance, otherwise Kruskal–Wallis
   analysis was used. Dunnett's t-test test was processed for two-by-two
   comparisons. P-values and correlation coefficients between differential
   bacteria and metabolites were calculated separately using spearman
   correlation analysis. Threshold at P < 0.05 and correlation
   coefficients > 0.8 were described in the microbiome-metabolite
   correlations. P < 0.05 was considered statistically significant.

Supplementary Information

   [127]Supplementary Information.^ (408.8KB, pdf)
   [128]Supplementary Table S1.^ (24.2KB, xlsx)
   [129]Supplementary Table S2.^ (9.2KB, xlsx)

Acknowledgements