INTRODUCTION
Acute pancreatitis (AP) is a major cause of gastrointestinal hospitalization and healthcare cost (1–3) with a higher mortality in those with severe AP (SAP) (4–6). Diet has been implicated as a potentially modifiable risk factor of AP; however, mechanistic insight is lacking (7,8). Diet high in animal protein and fat increases pancreatic proteases, leads to oxidative stress, and worsened pancreatic injury in animal studies (9–13). Conversely, diet low in animal protein and fat and rich in fruits, vegetables, and fish is beneficial to human health (14–21).
AP patients consuming diets rich in red meat are at 2.5 times higher risk of SAP (22). Diet can change the gut microbiome composition, and a previous study showed significant associations between diet and the abundance of hydrogen sulfide (H2S)–producing bacteria (23). By using endogenous and dietary sources of sulfur, enriched in animal protein and fat, sulfidogenic bacteria can increase H2S levels to induce inflammatory pathways (24,25). In fact, a higher H2S level has been detected in a small subset of AP patients (26). These data indicate that diet either by modulation of gut microbiota or through byproducts of gut microbial function might be modifying AP risk. Despite being a modifiable target, the interplay between diet and microbiome has not been studied prospectively in a diverse cohort of AP patients.
We hypothesized that a diet high in animal protein and fat and low in fiber is a disease-modifying factor increasing AP risk by modulating gut microbiome or its byproducts. In this case-control study, we assessed dietary intake and gut microbial composition and its function, as reflected by serum H2S and short-chain fatty acid (SCFA) levels among AP patients and controls. First, differences in consumption of diet among AP patients vs controls and the correlations between dietary variables and AP status were determined. Second, differences in gut microbial composition and abundance of sulfidogenic bacteria were examined. Next, differences in serum concentrations of SCFAs, the primary byproducts of gut microbial function, H2S, and C-reactive protein (CRP) were measured. Finally, using machine learning (ML) models, potential microbial targets associated with AP risk were explored.
METHODS
Study overview
An IRB approval (#2019-0910) was obtained. AP diagnosis was defined by at least 2 of the following: (i) abdominal pain consistent with AP, (ii) serum amylase or lipase ≥ 3 times the upper limit of normal, and (iii) features of AP on cross-sectional imaging. AP severity was based on the revised Atlanta classification (27). Exclusion criteria included age younger than 18 years, AP secondary to pancreas trauma, pain ≥7 days from the time of enrollment, chronic pancreatitis, pancreatic cancer, acute myocardial infarction, decompensated heart failure, decompensated liver cirrhosis, admission for infection, antibiotic use within 4 weeks before presentation, regular prebiotic or probiotic use, a gastrointestinal illness that would interfere with luminal absorption, and prior bariatric/malabsorptive surgery. To control for potential hospitalization-related gut microbiome changes (28–31), controls were recruited among hospitalized patients receiving care not related to pancreatic diseases. Controls were identified by a daily review of hospital admission records. The same exclusion criteria were applied to subjects in the control group. Patients who did not have a documented history of a pancreatic disease were approached and enrolled within 24 hours of their admission if they were interested and did not have an exclusion criterion.
Clinical data
Age, sex, race, ethnicity, body mass index (BMI), waist circumference, medical history, and alcohol, tobacco, and medication use data were collected.
Dietary data
An electronic food frequency questionnaire (VioCare, VioScreen version4) that queries habitual dietary intake (32) was administered within 48 hours of enrollment.
Biospecimens
Blood.
Samples were obtained after enrollment, processed for serum and plasma (33), and stored at −80°C.
SCFAs.
SCFAs (acetate, propionate, butyrate, and isovaleric acid) were isolated from 50 μL of serum (n = 49) and measured using gas chromatography-mass spectrometry (34–36).
Serum H2S.
Samples were treated with 1% zinc acetate to trap H2S using an adapted protocol (37) and assayed, and absorbance was read on a Tecan Infinite M200 ProMean.
Plasma CRP.
Samples were diluted 1:100 and measured in duplicates using a Human CRP Quantikine enzyme-linked immunosorbent assay (R&D systems) kit.
Stool sample collection and processing.
Stool samples were collected after enrollment and stored at −80°C. Subjects with no bowel movement before discharge were given a kit to collect the first stool sample at home with subsequent cold chain storage and transport. Microbial DNA was isolated using FastDNA SPIN Kit for Feces and quantified.
16S rDNA sequencing.
Genomic DNA was polymerase chain reaction amplified with primers targeting the V4 regions of ribosomal RNA. Amplicons were generated using a 2-stage polymerase chain reaction protocol (38),. Samples were pooled, purified, and loaded onto an Illumina MiniSeq. Forward and reverse reads were merged using PEAR (39) and trimmed (40) using cutadapt v1.18. Chimeric sequences were removed using the USEARCH algorithm (41). Amplicon sequence variants were identified using DADA2 (42).
Sequencing of the V4 region resulted in 3,162,840 raw reads that were demultiplexed, quality filtered to remove low abundant operational taxonomic units and singletons, and clustered using a 97% similarity threshold through QIIME 1.9.1. Quality filtering and rarefaction to 252 (max = 390) sequences per sample resulted in 3,162,801 reads from 36 samples (20 AP and 16 control subjects), with a sample mean of 87,856 reads.
Bioinformatics processing.
For 16S rDNA analysis, paired-end reads were generated and joined followed by quality filtering (43) and demultiplexing in QIIME 1.9.1 (44), and exact sequence variants (ESVs) were identified using DeBlur pipeline (45). The final biom file was rarefied and analyzed in R package (46). Microbial alpha diversity was calculated by the Shannon index using ESVs in 16S rRNA amplicon sequences.
Statistical analysis.
For continuous variables, the distribution of normality was evaluated. Parametric or nonparametric tests were chosen, as appropriate. Density variables were created for key nutrients and food items per 1,000 calories. Differences between the 2 groups were analyzed using the Wilcoxon rank-sum test or t test for continuous variables and the χ2 or Fisher exact test for dichotomous variables. More than 2 groups were analyzed using 1-way ANOVA (47) or the Kruskal-Wallis test. Statistical analyses were performed using SPSS (version 27, IBM SPSS Statistics), and figures were developed using Prism (GraphPad, version 9.2.0). All statistical tests were 2-sided, and P values <0.05 were considered statistically significant.
MetagenomeSeq (48) was applied to test the differential abundance of microbial groups. All ESV counts were normalized, and a zero-inflated Gaussian mixture model was used to validate the zero counts from microbial composition. The results were corrected for diabetes status.
Linear discriminant analysis (LDA) effect size (LEfSe) (49) was applied to identify the 16S rDNA biomarker to differentiate between AP and control subjects. The 1-way nonparametric Kruskal-Wallis sum rank test was performed. Then, LDA was used to estimate the effect size (log10) of each differentially abundant genus for those genera with effect sizes >2. A phylogenetic tree of differentially abundant microbial groups were plotted in a cladogram.
Random Forest classifiers (50), under R (version 4.04), were trained to discriminate AP subjects from controls using microbiome genera along with diabetes status, age, sex, and serum acetate concentrations in a study subgroup. The ntree parameter was set to 500, and mtry was set to 11 to ensure the diversity of the trees and reduce the greediness of learning. Leave-one-out cross-validation procedure was used, and model performance was evaluated based on receiver operating characteristic (ROC) curves. For model validation, the target model was trained 100 times with 25 significant features, feature importance > 0.2.
RESULTS
Baseline comparison of study groups
One hundred subjects (AP = 54, control = 46) were enrolled. Clinical and sociodemographic differences are summarized in Table 1. The average time to enrollment from onset of pain was 2.5 days for AP subjects. The most common AP etiologies were alcohol (42.6%) and gallstones (25.9%), followed by idiopathic AP (11.1%), other causes (11.1%), and hypertriglyceridemia (9.3%) (Figure 1a). AP severity was classified as mild, moderately severe, and severe (79.6%, 16.7%, and 3.7%, respectively) (Figure 1b). Last, 35.2% of AP subjects had a history of AP. The following were the primary admission diagnoses for study participants in the control group: cardiovascular (chest pain, palpitations, syncope, atrial fibrillation, peripheral arterial disease, and hypertension), musculoskeletal (low back pain, rhabdomyolysis, hand pain, shoulder pain, and osteoarthritis), pulmonary (asthma, shortness of breath, and pneumothorax), neurology (vertigo, cervical stenosis, headache, and seizure), endocrinology (type 2 diabetes mellitus and hypoglycemia), obstetrics and gynecology (uterine fibroids and abnormal uterine bleeding), gastrointestinal (gastroesophageal reflux disease and anal fissure), dermatology (rash), otolaryngology (chronic rhinosinusitis), ophthalmology (eye swelling), and rheumatology (gout).
Table 1. -
Differences in clinical and sociodemographic characteristics of control and acute pancreatitis subjects
|
Control (N = 46) |
Acute pancreatitis (N = 54) |
P value |
| Age (yrs) |
45.67 ± 14.8 |
45.65 ± 14.49 |
0.993 |
| Sex (female/male) (%) |
58.7/41.3 |
46.3/53.7 |
0.338 |
| Race, AA/White (%) |
75.6/24.4 |
38.9/61.1 |
0.001
|
| Ethnicity, Hispanic/Non-Hispanic (%) |
15.2/84.8 |
42.6/57.4 |
0.006
|
| Body mass index (kg/m2) |
33.7 ± 8.19 |
30.4 ± 7.70 |
0.038
|
| Waist circumference (cm) |
74.6 ± 27.2 |
85.3 ± 24.6 |
0.045
|
| Medication use and smoking |
|  Statin use (%) |
25.0 |
29.6 |
1 |
|  Nonsteroidal anti-inflammatory drug use (%) |
23.9 |
38.9 |
0.166 |
|  Proton pump inhibitor use (%) |
28.3 |
31.5 |
0.895 |
|  Current smoking (%) |
28.3 |
29.6 |
1 |
| Medical history |
|  Peptic ulcer (%) |
4.3 |
20.4 |
0.038
|
|  Gallstones (%) |
4.3 |
25.9 |
0.008
|
|  Celiac disease (%) |
0 |
0 |
1 |
|  Diabetes (%) |
30.4 |
20.4 |
0.354 |
|  Thyroid disease (%) |
8.7 |
3.7 |
0.532 |
|  Chronic kidney disease (%) |
2.2 |
5.6 |
0.728 |
|  Myocardial infarction (%) |
6.5 |
3.7 |
0.854 |
|  Hypertension (%) |
63.0 |
48.1 |
0.197 |
|  Peripheral vascular disease (%) |
2.2 |
0 |
0.936 |
|  Asthma (%) |
30.4 |
11.1 |
0.031
|
|  Emphysema/COPD (%) |
2.2 |
7.4 |
0.461 |
|  Charlson comorbidity index score |
1.19 ± 1.81 |
1.11 ± 1.49 |
0.82 |
AA, African American; COPD, chronic obstructive pulmonary disease.
;)
Figure 1.: The etiology, disease severity, and history of previous attacks in the acute pancreatitis cohort. (a) The most common etiologies were alcohol use followed by gallstones. (b) Approximately 20% of AP patients had moderately severe or severe disease. AP, acute pancreatitis.
Acute pancreatitis patients consume lower amounts of beneficial foods and nutrients
Although the mean total carbohydrate, total protein, and total fat per 1,000 kcal intake were similar between 2 groups, AP subjects had a significantly lower consumption of certain beneficial polyunsaturated fatty acids (PUFAs) (Table 2). Furthermore, AP subjects had a significantly lower intake of protein from seafood/fish, whole grains, daily servings of fish, vitamin D3, and total transfatty acids per 1,000 kcal (Table 2). In addition, using the Spearman rank correlation and after age, sex, and race adjustment, significant correlations were identified between AP status and key dietary variables (see Supplementary Figure 1, https://links.lww.com/CTG/A943, https://links.lww.com/CTG/A945).
Table 2. -
Differences in habitual dietary intake among acute pancreatitis and control subjects (per 1,000 kcal)
| Dietary variables |
Acute pancreatitis (N = 48) |
Control (N = 42) |
P value |
| Total dietary fat (g) |
37.5 (10.1) |
38.5 (7.61) |
0.586 |
|  Total polyunsaturated fatty acids (PUFA) (g) |
7.89 (2.64) |
8.59 (2.26) |
0.181 |
|   Arachidonic acid (PUFA 20:4) (g) |
0.058 (0.038) |
0.076 (0.037) |
0.013
|
|   Eicosapentaenoic acid (PUFA 20:5) (g) |
0.011 (0.013) |
0.016 (0.012) |
0.003
|
|   Docosapentaenoic acid (PUFA 22:5) (g) |
0.008 (0.006) |
0.011 (0.006) |
0.004
|
|   Docosahexaenoic acid (PUFA 22:6) (g) |
0.023 (0.027) |
0.036 (0.026) |
0.003
|
|  Total monounsaturated fatty acids (MUFA) (g) |
13.94 (5.06) |
13.58 (3.71) |
0.923 |
|  Total transfatty acids (g) |
1.19 (0.62) |
1.54 (0.58) |
0.008
|
|  Total saturated fatty acids (g) |
12.2 (4.00) |
12.9 (3.04) |
0.349 |
| Total carbohydrate (g) |
117.9 (26.9) |
126.1 (18.9) |
0.106 |
| Total protein (g) |
34.7 (8.19) |
36.4 (9.15) |
0.371 |
|  Protein from seafood/fish (oz) |
0.046 (0.092) |
0.067 (0.086) |
0.016
|
| Vitamin D3 (mcg) |
2.30 (1.82) |
2.95 (1.85) |
0.040
|
| Whole grains (oz) |
0.49 (0.65) |
0.76 (0.93) |
0.050
|
| Daily servings of fish |
0.24 (0.29) |
0.44 (0.48) |
0.022
|
| Daily servings of alcohol |
0.75 (1.12) |
0.21 (0.52) |
0.106 |
Dietary variable values are adjusted per 1,000 kcal. Statistically significant P values are bolded.
Acute pancreatitis patients have lower gut microbiome diversity
Diabetes is a known variable that influences gut microbiome diversity. To evaluate gut microbiome diversity independent of diabetes status, we examined the differences in gut microbiome diversity between controls without diabetes, controls with diabetes, AP subjects without diabetes, and AP subjects with diabetes. Gut bacterial alpha diversity measured by the Shannon index (46) was lower among AP subjects (adjusted for race, ethnicity, age, and sex), and this difference was more significant among AP subjects with diabetes (P = 0.021; Figure 2a–c). The Shannon index remained significantly different after additional adjustment was completed for BMI (P = 0.04).
Figure 2.: Acute pancreatitis patients have decreased gut microbiome alpha diversity. (a) Gut microbiome alpha diversity was lower in subjects with AP, and this difference was more pronounced in AP patients with diabetes (P = 0.021). (b) Relative abundance of genera among controls, AP patients without diabetes, and AP patients with diabetes. (c) Representation of various genera that were identified during 16S sequencing among study groups. AP, acute pancreatitis.
AP patients have distinct gut microbiota composition and a higher relative abundance of some sulfidogenic bacteria. MetagenomeSeq identified the following genera to be more abundant in AP subjects compared with that in controls: Granulicatella sp., Veillonella sp., Klebsiella sp., Enterococcus sp., Catenisphaera sp., Actinomyces sp., Haemophilus sp., Victivallis sp., Barnesiella sp., Rothia sp., Gemella sp., Cloacibacillus sp., Morganella sp., Megamonas sp., Olsenella sp., Lachnospiraceae UCG-004, Coprobacter sp., Prevotella 2 sp. (Figure 3a). Among these genera, some sulfidogenic bacteria (Veillonella sp., Klebsiella sp., Enterococcus sp., Actinomyces sp., Haemophilus sp., Rothia sp., Gemella sp., Morganella sp., and Prevotella 2 sp.) were noted to be at a significantly higher abundance in AP compared with that in the control group (Figure 3a).
Figure 3.: Differential abundance of genera between acute pancreatitis and control subjects. (a) Linear discriminant analysis (LDA) showing differential abundance of genera associated with AP vs control status. (b) Specific genera associated with acute pancreatitis status based on LDA effect size (LEfSe) method. (c) Cladogram showing distinct differences and associations among bacteria in the AP vs control groups. AP, acute pancreatitis.
LEfSe (49) further identified that Veillonella sp. and Haemophilus sp. were strongly associated with AP, and 6 other genera were associated with control group (Figure 3b). In addition, the phylogenetic tree of microbial groups estimated and represented in a Cladogram shows that bacteria associated with AP had distinct relations and separated from bacteria associated with the control group (Figure 3c).
Acute pancreatitis patients have significantly higher serum acetic acid, H2S, and CRP concentrations
While no significant differences in serum propionate, butyrate, and isovaleric acid levels between AP and control groups were observed, AP group had significantly higher acetate levels (Figure 4; P < 0.001). This difference remained significant after adjustment for age, sex, BMI, lactated Ringer infusion, alcohol intake, and ethnicity (P = 0.0012). Race adjustment showed this difference to remain significant in African Americans (P = 0.003) but not in White individuals.
Figure 4.: Serum concentration of acetic acid in acute pancreatitis and control subjects. AP subjects had significantly higher serum acetic acid levels compared with controls (P < 0.001). AP, acute pancreatitis.
Patients with AP had significantly higher H2S concentrations compared with controls (3.17 ± 2.65 vs 1.87 ± 1.15 μM, P = 0.043). Similarly, subjects with AP had significantly higher CRP concentrations compared with controls (30.9 ± 18.5 vs 13.5 ± 14.8 μg/mL, P < 0.001).
Gut microbiome as a key component for determination of AP status
First, we assessed the predictive performance of model trained using the 128 taxa, acetic acid, age, sex, and diabetes status. Race and BMI reduced the overall model performance and were removed from training data. Overall, 89% of the area was under the ROC curve, which indicated the predictive ability of the model to distinguish between AP and control subjects (see Supplementary Figure 2, https://links.lww.com/CTG/A944, https://links.lww.com/CTG/A945). A final model (Figure 5a) was trained on the top 25 features (24 taxa and acetic acid) ranked by random forest with feature importance greater than 0.2 (Figure 5c). Overall, 96% of the area was under the ROC curve (Figure 5a), which outperformed the previous model and indicated a significantly higher predictive ability of the model to distinguish between AP and control subjects. The target model has significantly higher AUC than the null model (P = 2.2e-16, 95% confidence interval [0.53–0.57]).
Figure 5.: Machine learning model to distinguish acute pancreatitis subjects from controls. (a) The receiver operating characteristics curve showing 96% predictive ability of the model to distinguish between AP and control subjects based on 24 taxa, combined with acetic acid level. (b) The model validation result showing the 95% confidence interval for the differences between the target and the null model. Target model was trained using selected features with importance higher than 0.2, shown in panel (c). Null model was trained using features with importance lower than 0.2. (c) Top 25 features ranked by random forest with feature importance greater than 0.2. AP, acute pancreatitis.
DISCUSSION
In this case-control study, we found that AP patients' diet is not only low in beneficial nutrients such as key PUFAs, omega-3, whole grains, and vitamin D but also AP patients have lower gut microbiome diversity, which is more pronounced in AP patients with preexisting diabetes. Of importance, 16S microbiome analysis showed that AP patients have a higher abundance of some of the sulfidogenic bacteria, known inducers of inflammation in the gut. Furthermore, serum acetate and H2S levels were significantly higher in the AP group. These data suggest that diet and gut microbiota through their end metabolites might be playing an important role in AP risk. In fact, using gut microbiome and key clinical data in ML algorithm, we had 96% predictive ability to differentiate AP subjects from controls.
AP etiology and severity data in our study are in line with published literature (51,52). The dietary data revealed a decreased intake of fish, vitamin D, whole grains, and beneficial PUFAs (53–57). AP patients are more likely to develop vitamin D deficiency (58), which has been linked to immune and glycemic dysfunction (59). Whole grains are rich sources of phytochemicals, dietary fiber, complex carbohydrates, and vitamins. We observed a decreased consumption of PUFA 20:5, PUFA 22:5, PUFA 20:6, protein from seafood/fish, whole grains, daily servings of fish, and vitamin D3 in AP patients. Consumption of a diet low in beneficial nutrients in the AP group indicates that dietary intake could be contributing to decreased gut microbiome diversity in AP. Although 65% of AP patients were started on oral intake by day 1, diet survey used in our study collects habitual dietary intake data over the past 3 months, showing existence of dietary intake differences between 2 groups before AP attack. In addition, stool samples were already collected by day 1 in 70% of AP patients by sampling of the first bowel movements during hospitalization. Early collection of stool specimens along with sampling of first bowel movements indicates that gut microbiome data in the AP group reflect the existing gut microbiome composition before or during AP attack. Stool sample collection rate at day 1 was 100% in the control group. The lower rate of stool sample collection at day 1 in the AP group (70%) is attributed to a decreased oral intake before admission due to abdominal pain and nothing by mouth status after AP admission.
Increased gut microbiota diversity is beneficial for human health (60) and immune homeostasis (61). Two studies described decrease in gut microbiome diversity in AP (62,63) in humans, and less is known about underlying mechanisms. We also identified decreased gut microbiota diversity in the AP group, yet a detailed phenotyping of AP patients, diet, and SCFA data provided further mechanistic insight. For example, alpha diversity was low among AP subjects, and this was most pronounced in AP subjects with diabetes. Furthermore, patients with a history of bariatric/malabsorptive surgery were excluded. Among study participants with available gut microbiome data, only 1 AP patient had ileus, and no participants had documented history of dysmotility. The difference in gut microbiome diversity remained statistically significant once adjustment was completed for the presence of ileus (P = 0.031). In addition, there was no statistically significant difference in the rate of antibiotic use during hospitalization between AP (16.6%) and control (24%) groups (P = 0.45). Therefore, dysbiosis seen in the AP group really reflects the decreased gut microbiome diversity.
Our 16S microbiome analysis demonstrated an increased abundance of some of the sulfidogenic bacteria in the AP group. LDA analysis (Figure 3.B) showed that Veillonella sp. and Hemophilus sp. were significantly elevated in the AP group. Veillonella sp., a Gram-negative sulfidogenic bacterium that resides in the gastrointestinal tract and oral cavity (64) can produce H2S from L-cysteine, which is abundantly present in meat (65,66). Of interest, Veillonella sp. has also been associated with gestational diabetes, complications of diabetes (67), and pancreatic adenocarcinoma (68). Haemophilus sp. is a sulfur-reducing commensal bacteria of the human oropharynx and has been associated with periodontal disease and pancreatic abscess (69) and severe alcoholic hepatitis (70).
Our study also demonstrated a higher abundance of Granulicatella sp., which have been associated with pancreatic cancer (71), and Klebsiella pneumoniae sp., a sulfidogenic bacterium, which is a common cause of infection in intensive care units (72,73). Enterococcus, a cysteine-specific H2S producer (74) that has been identified in patients with SAP, infected pancreatic necrosis, and organ failure (75,76); and Catenisphaera sp. which have been implicated in tight junction and basement membrane damage in the intestine (77) were also more abundant in the AP group. Actinomyces sp., a commensal sulfidogenic bacterium (78) associated with inflammatory diseases (78,79), and Rothia sp., H2S producing bacteria (80) detected in higher abundance in pancreatic cancer (81), were also more abundant in our AP cohort. Similarly, the following sulfidogenic bacteria were also at a higher abundance in the AP group: Gemella sp., producer of volatile sulfur compounds (82), Morganella sp., a pathogen for intra-abdominal abscess (83) formation, and Prevotella 2 sp., H2S producer associated with an increased risk of inflammatory diseases and insulin resistance (84). A significantly higher abundance of acetate-producing bacteria (Catenisphaera sp., Victivallis sp., Barnesiella sp., Cloacibacillus sp., Megamonas sp., Olsenella sp., and Coprobacter sp.) in the AP group may also explain higher acetate levels in the AP group.
SCFAs are considered metabolically beneficial for human health. In 16S analysis, the control group (Figure 3a) had a higher abundance of certain SCFA-producing bacteria including Prevotellaceae NK3B31 (85), Ruminococcaceae UCG-010 (86), and Coprococcus 2 sp. (87). In addition, Fusicatenibacter sp., SCFA producer associated with decreased intestinal inflammation (88,89), and Blautia sp., a potential target for obesity and diabetes due to its anti-inflammatory features (90–92), were also more abundant in the control group. Lachnospiraceae NK3A20, butyric acid producer that maintains the colonic epithelial barrier (93), and Lactobacillus sp., which alleviate the magnitude of injury in AP in humans and animals (94–96), were also more abundant in the control group.
Production of inflammatory H2S (23) by sulfidogenic bacteria may contribute to AP risk, and significantly higher serum H2S concentrations in our AP group is in line with this hypothesis. CRP concentrations are elevated in AP and proposed as a prediction marker for AP severity (97,98). Significantly higher CRP levels in AP subjects in our study is also in line with these reports.
AP patients with severe disease are at a higher risk of exocrine and endocrine insufficiencies, and development of accurate prediction models is a priority for the field (99,100). We used an ML model that differentiates AP patients from controls based on gut microbiome profile and key clinical characteristics with 96% predictive ability. Our novel model consolidates the importance of key microbial genera such Veillonella sp. and Haemophilus sp. and increased levels of acetic acid, end product of alcohol degradation, in AP patients. Presence of a significant correlation between alcohol intake and AP status in our cohort, predominance of alcohol-induced AP (43%), and abundance of acetic acid–producing genera (Catenisphaera sp., Victivallis sp., Barnesiella sp., Cloacibacillus sp., Megamonas sp., Olsenella sp., and Coprobacter sp.) in AP patients can explain the higher acetic acid levels in the AP group. Further development and validation of this ML model in a large clinical cohort can provide opportunities to identify patients at risk of recurrent AP, SAP, and AP-related endocrine and exocrine insufficiencies. Potential preventive measures and therapeutic strategies can be developed and implemented in these high-risk groups to decrease cycles of inflammatory injury and AP-related disease burden.
Our study has several limitations. First, food frequency questionnaire is subject to recall bias. However, we used a validated survey that limits such bias. Second, we were not able to complete subgroup analysis based on AP etiologies due to relatively small sample size. Nevertheless, we identified novel gut microbiota signatures between AP and control subjects and plan to investigate the role of AP etiologies in these differences in a larger cohort study in future. Third, hospital admission may have an impact on gut microbiome composition. We have controlled for this by enrolling AP and control subjects from hospital and excluding those who were on antibiotics or prebiotics/probiotics. Next, our 16S microbiome analysis was based on a study subgroup, and ML model should be trained through different cohorts to prevent overfitting and for further validation. However, our findings provide framework for larger studies in which 16S microbiome and metabolomic analyses could be completed for the entire study. Last, given our sample size, replicating these findings is necessary.
In summary, using a validated diet questionnaire, we have shown that AP patients consume diets low in beneficial nutrients. The presence of decreased gut microbiome diversity in our AP group can be partially explained by consumption of diet that is not metabolically beneficial. An increased abundance of bacteria that produce SCFAs in the control and sulfidogenic bacteria in the AP group along with higher H2S concentration indicates that diet, gut microbiota, and their end metabolites might play a role in AP development. On replication in larger clinical cohorts, these findings may be translated into clinical practice through diet or microbiome-based prevention or treatment strategies and decrease AP-related morbidity and mortality.
CONFLICTS OF INTEREST
Guarantor of the article: Cemal Yazici, MD, MS.
Specific author contributions: C.Y. and E.M.: conception and design. S.T., K.C., H.A.R., C.Y., Y.H., P.S., Y.D.: data organization and analysis. C.Y., Y.D., E.M., B.L.: interpretation of data and drafting the article. C.Y., B.B., P.G., R.G., K.K.D., G.P., L.T.-H., Y.D., E.M., B.L.: revising of the article for critically important intellectual content. All authors approved the final version of this manuscript.
Financial support: This project was supported by the National Institute of Health (NIH) through grant number KL2TR002002, which provided support for C.Y. Additional support was provided to C.Y. and B.T.L. by National Institute of Diabetes and Digestive and Kidney Disease (NIDDK) through grant number U01DK127378 and to K.C. through grant number 3U01DK127378-02S1. B.T.L. was also supported by grants NIH R01DK104927-01A1, P30DK020595, and VA merit 1I01BX003382-01-A1. The content is solely the responsibility of the authors, was independent of funding support, and does not necessarily represent the official views of the NIH, and the study sponsor did not participate in the study design, data collection, analysis or interpretation.
Potential competing interests: None to report.
IRB approval statement: This study was approved by the Institutional Review Board (IRB) of the University of Illinois Chicago (IRB approval #2019-0910).Study Highlights
WHAT IS KNOWN
- ✓ Diet has been implicated as a risk factor in acute pancreatitis.
- ✓ Diet can modify the diversity of the gut microbiome.
- ✓ Decreased gut microbiome diversity has been associated with acute pancreatitis.
- ✓ The association between diet and gut microbiome and their relationship to microbial metabolites has not been studied in acute pancreatitis.
WHAT IS NEW HERE
- ✓ Patients with acute pancreatitis consume lower amounts of beneficial nutrients and have decreased gut microbiome diversity.
- ✓ Patients with acute pancreatitis have a higher abundance of sulfidogenic bacteria and higher serum hydrogen sulfide and acetate concentrations.
- ✓ Gut microbiome composition along with serum acetate levels associate with acute pancreatitis status in machine learning models.
REFERENCES
1. Peery AF, Crockett SD, Murphy CC, et al. Burden and cost of gastrointestinal, liver, and pancreatic diseases in the United States: Update 2018. Gastroenterology 2019;156(1):254–72.e11.
2. Yadav D, Lowenfels AB. The epidemiology of pancreatitis and pancreatic cancer. Gastroenterology 2013;144(6):1252–61.
3. Krishna SG, Kamboj AK, Hart PA, et al. The changing epidemiology of acute pancreatitis hospitalizations: A decade of trends and the impact of chronic pancreatitis. Pancreas 2017;46(4):482–8.
4. Peery AF, Dellon ES, Lund J, et al. Burden of gastrointestinal disease in the United States: 2012 update. Gastroenterology 2012;143(5):1179–87.e3.
5. Whitcomb DC. Acute pancreatitis. N Engl J Med 2006;354(20):2142–50.
6. McNabb-Baltar J, Ravi P, Isabwe GA, et al. A population-based assessment of the burden of acute pancreatitis in the United States. Pancreas 2014;43(5):687–91.
7. Setiawan VW, Pandol SJ, Porcel J, et al. Dietary factors reduce risk of acute pancreatitis in a large multiethnic cohort. Clin Gastroenterol Hepatol 2017;15(2):257–65.e3.
8. Yang H, Wang L, Shi YH, et al. Risk factors of acute pancreatitis in the elderly Chinese population: A population-based cross-sectional study. J Dig Dis 2014;15(9):501–7.
9. Whitcomb DC. Genetics and alcohol: A lethal combination in pancreatic disease? Alcohol Clin Exp Res 2011;35(5):838–42.
10. Wilson JS, Bernstein L, McDonald C, et al. Diet and drinking habits in relation to the development of alcoholic pancreatitis. Gut 1985;26(9):882–7.
11. Grossman MI, Greengard H, Ivy AC. The effect of dietary composition on pancreatic enzymes. Am J Physiology Legacy Content 1943;138(4):676–82.
12. Brian Haig TH. Experimental pancreatitis intensified by a high fat diet. Surg Gynecol Obstet 1970;131(5):914–8.
13. Lindsay S, Entenman C, Chaikoff IL. Pancreatitis accompanying hepatic disease in dogs fed a high fat, low protein diet. Arch Pathol 1948;45:635–8.
14. Caughey GE, Mantzioris E, Gibson RA, et al. The effect on human tumor necrosis factor alpha and interleukin 1 beta production of diets enriched in n-3 fatty acids from vegetable oil or fish oil. Am J Clin Nutr 1996;63(1):116–22.
15. Nalsen C, Vessby B, Berglund L, et al. Dietary (n-3) fatty acids reduce plasma F2-isoprostanes but not prostaglandin F2α in healthy humans. J Nutr 2006;136(5):1222–8.
16. Mas E, Woodman RJ, Burke V, et al. The omega-3 fatty acids EPA and DHA decrease plasma F(2)-isoprostanes: Results from two placebo-controlled interventions. Free Radic Res 2010;44(9):983–90.
17. Endres S, Ghorbani R, Kelley VE, et al. The effect of dietary supplementation with n-3 polyunsaturated fatty acids on the synthesis of interleukin-1 and tumor necrosis factor by mononuclear cells. N Engl J Med 1989;320(5):265–71.
18. Ananthakrishnan AN, Khalili H, Konijeti GG, et al. Long-term intake of dietary fat and risk of ulcerative colitis and Crohn's disease. Gut 2014;63(5):776–84.
19. Ananthakrishnan AN, Khalili H, Konijeti GG, et al. A prospective study of long-term intake of dietary fiber and risk of Crohn's disease and ulcerative colitis. Gastroenterology 2013;145(5):970–7.
20. Chan DS, Lau R, Aune D, et al. Red and processed meat and colorectal cancer incidence: meta-analysis of prospective studies. PLoS One 2011;6(6):e20456.
21. Di Giuseppe D, Wallin A, Bottai M, et al. Long-term intake of dietary long-chain n-3 polyunsaturated fatty acids and risk of rheumatoid arthritis: A prospective cohort study of women. Ann Rheum Dis 2014;73(11):1949–53.
22. Dugum M, Gougol A, Paragomi P, et al. Association of dietary habits with severity of acute pancreatitis. Curr Dev Nutr 2018;2(12):nzy075.
23. Yazici C, Wolf PG, Kim H, et al. Race-dependent association of sulfidogenic bacteria with colorectal cancer. Gut 2017;66(11):1983–94.
24. Deplancke B, Gaskins HR. Hydrogen sulfide induces serum-independent cell cycle entry in nontransformed rat intestinal epithelial cells. FASEB J 2003;17(10):1310–2.
25. Attene-Ramos MS, Nava GM, Muellner MG, et al. DNA damage and toxicogenomic analyses of hydrogen sulfide in human intestinal epithelial FHs 74 Int cells. Environ Mol Mutagen 2010;51(4):304–14.
26. Thangaraj KR, Priyadarshini SJ, Qureshi IN, et al. Plasma citrulline, glycans, and hydrogen sulfide in patients with acute pancreatitis: Possible markers of intestinal damage. Pancreas 2016;45(6):e27–29.
27. Banks PA, Bollen TL, Dervenis C, et al. Classification of acute pancreatitis: 2012: Revision of the Atlanta classification and definitions by international consensus. Gut 2013;62(1):102–11.
28. Vincent C, Miller MA, Edens TJ, et al. Bloom and bust: Intestinal microbiota dynamics in response to hospital exposures and Clostridium difficile colonization or infection. Microbiome 2016;4(1):12.
29. Poza M, Gayoso C, Gomez MJ, et al. Exploring bacterial diversity in hospital environments by GS-FLX Titanium pyrosequencing. PLoS One 2012;7(8):e44105.
30. Lauretani F, Longobucco Y, Maggio M. Gut microbiota and motoric-cognitive frailty in hospitalized older persons. Curr Opin Clin Nutr Metab Care 2021;24(3):209–15.
31. Ghosh TS, Das M, Jeffery IB, et al. Adjusting for age improves identification of gut microbiome alterations in multiple diseases. Elife 2020;9:e50240.
32. Kristal AR, Kolar AS, Fisher JL, et al. Evaluation of web-based, self-administered, graphical food frequency questionnaire. J Acad Nutr Diet 2014;114(4):613–21.
33. Fisher WE, Cruz-Monserrate Z, McElhany AL, et al. Standard operating procedures for biospecimen collection, processing, and storage: From the consortium for the study of chronic pancreatitis, diabetes, and pancreatic cancer. Pancreas 2018;47(10):1213–21.
34. Moreau NM, Goupry SM, Antignac JP, et al. Simultaneous measurement of plasma concentrations and 13C-enrichment of short-chain fatty acids, lactic acid and ketone bodies by gas chromatography coupled to mass spectrometry. J Chromatogr B Analyt Technol Biomed Life Sci 2003;784(2):395–403.
35. Richardson AJ, Calder AG, Stewart CS, et al. Simultaneous determination of volatile and non-volatile acidic fermentation products of anaerobes by capillary gas chromatography. Lett Appl Microbiol 1989;9(1):5–8.
36. Lie L, Brown L, Forrester TE, et al. The association of dietary fiber intake with cardiometabolic risk in four countries across the epidemiologic transition. Nutrients 2018;10(5):628.
37. Ahmad FU, Sattar MA, Rathore HA, et al. Exogenous hydrogen sulfide (H2S) reduces blood pressure and prevents the progression of diabetic nephropathy in spontaneously hypertensive rats. Ren Fail 2012;34(2):203–10.
38. Naqib A, Poggi S, Wang W, et al. Making and sequencing heavily multiplexed, high-throughput 16S ribosomal RNA gene amplicon libraries using a flexible, two-stage PCR protocol. Methods Mol Biol 2018;1783:149–69.
39. Zhang J, Kobert K, Flouri T, et al. Pear: A fast and accurate Illumina paired-end reAd mergeR. Bioinformatics 2014;30(5):614–20.
40. Martin M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnetjournal 2011;17(1):10.
41. Edgar RC. Search and clustering orders of magnitude faster than BLAST. Bioinformatics 2010;26(19):2460–1.
42. Callahan BJ, McMurdie PJ, Rosen MJ, et al. DADA2: High-resolution sample inference from Illumina amplicon data. Nat Methods 2016;13(7):581–3.
43. Bokulich NA, Subramanian S, Faith JJ, et al. Quality-filtering vastly improves diversity estimates from Illumina amplicon sequencing. Nat Methods 2013;10(1):57–9.
44. Caporaso JG, Kuczynski J, Stombaugh J, et al. QIIME allows analysis of high-throughput community sequencing data. Nat Methods 2010;7(5):335–6.
45. Amir A, McDonald D, Navas-Molina JA, et al. Deblur rapidly resolves single-nucleotide community sequence patterns. mSystems 2017;2(2):e00191-16.
46. McMurdie PJ, Holmes S. phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS One 2013;8(4):e61217.
47. Kaufmann J, Schering A. Analysis of variance ANOVA. 2014. doi: 10.1002/9781118445112.stat06938.
48. Paulson JN, Stine OC, Bravo HC, et al. Differential abundance analysis for microbial marker-gene surveys. Nat Methods 2013;10(12):1200–2.
49. Segata N, Izard J, Waldron L, et al. Metagenomic biomarker discovery and explanation. Genome Biol 2011;12(6):R60.
50. Breiman L. Random forests. Machine Learn 2001;45(1):5–32.
51. Arteel GE, Singhvi A, Feldman R, et al. Coexistent alcohol-related liver disease and alcohol-related pancreatitis: Analysis of a large health care system cohort. Dig Dis Sci 2021;67(6):2543–51.
52. Weiss FU, Laemmerhirt F, Lerch MM. Acute pancreatitis: Genetic risk and clinical implications. J Clin Med 2021;10(2):190.
53. Fu Y, Wang Y, Gao H, et al. Associations among dietary omega-3 polyunsaturated fatty acids, the gut microbiota, and intestinal immunity. Mediators Inflamm 2021;2021:1–11.
54. Parikh M, Maddaford TG, Austria JA, et al. Dietary flaxseed as a strategy for improving human health. Nutrients 2019;11(5):1171.
55. Cholewski M, Tomczykowa M, Tomczyk M. A comprehensive review of chemistry, sources and bioavailability of omega-3 fatty acids. Nutrients 2018;10(11):1662.
56. Basson AR, Chen C, Sagl F, et al. Regulation of intestinal inflammation by dietary fats. Front Immunol 2020;11:604989.
57. Mensink RP, World Health O. Effects of Saturated Fatty Acids on Serum Lipids and Lipoproteins: A Systematic Review and Regression Analysis. World Health Organization: Geneva, Switzerland, 2016.
58. Abou Saleh M, Alkhayyat M, Mansoor E, et al. The risk of vitamin D deficiency, osteoporosis, and fractures in acute pancreatitis. Pancreas 2020;49(5):629–33.
59. Pilz S, Gaksch M, Kienreich K, et al. Effects of vitamin D on blood pressure and cardiovascular risk factors: A randomized controlled trial. Hypertension 2015;65(6):1195–201.
60. Lau A, Tan LTH, Ab Mutalib NS, et al. The chemistry of gut microbiome in health and diseases. Prog Microbes Mol Biol 2021;4(1).
61. Ley RE, Peterson DA, Gordon JI. Ecological and evolutionary forces shaping microbial diversity in the human intestine. Cell 2006;124(4):837–48.
62. Brubaker L, Luu S, Hoffman K, et al. Microbiome changes associated with acute and chronic pancreatitis: A systematic review. Pancreatology 2021;21(1):1–14.
63. Zhu Y, He C, Li X, et al. Gut microbiota dysbiosis worsens the severity of acute pancreatitis in patients and mice. J Gastroenterol 2019;54(4):347–58.
64. Shih CT, Yeh YT, Lin CC, et al. Akkermansia muciniphila is negatively correlated with hemoglobin A1c in refractory diabetes. Microorganisms 2020;8(9):1360.
65. Washio J, Shimada Y, Yamada M, et al. Effects of pH and lactate on hydrogen sulfide production by oral Veillonella spp. Appl Environ Microbiol 2014;80(14):4184–8.
66. Feng Y, Stams AJM, de Vos WM, et al. Enrichment of sulfidogenic bacteria from the human intestinal tract. FEMS Microbiol Lett 2017;364(4).
67. Al-Otaibi FE, Al-Mohizea MM. Non-vertebral Veillonella species septicemia and osteomyelitis in a patient with diabetes: A case report and review of the literature. J Med Case Rep 2014;8(1):365.
68. Pandya G, Kirtonia A, Singh A, et al. A comprehensive review of the multifaceted role of the microbiota in human pancreatic carcinoma. Semin Cancer Biol 2022;86(Pt 3):682–92.
69. Carson HJ, Rezmer S, Belli J. Haemophilus segnis cholecystitis: A case report and literature review. J Infect 1997;35(1):85–6.
70. Ciocan D, Rebours V, Voican CS, et al. Characterization of intestinal microbiota in alcoholic patients with and without alcoholic hepatitis or chronic alcoholic pancreatitis. Sci Rep 2018;8(1):4822.
71. Memba R, Duggan SN, Ni Chonchubhair HM, et al. The potential role of gut microbiota in pancreatic disease: A systematic review. Pancreatology 2017;17(6):867–74.
72. Arteaga-Livias K, Pinzas-Acosta K, Perez-Abad L, et al. A multidrug-resistant Klebsiella pneumoniae outbreak in a Peruvian hospital: Another threat from the COVID-19 pandemic. Infect Control Hosp Epidemiol 2021;43:267–8.
73. Braunstein H, Tomasulo M. Indole-positive strains of Klebsiella pneumoniae producing hydrogen sulfide in iron-agar slants. Am J Clin Pathol 1976;65(5):702–5.
74. Yoshida A, Yoshimura M, Ohara N, et al. Hydrogen sulfide production from cysteine and homocysteine by periodontal and oral bacteria. J Periodontol 2009;80(11):1845–51.
75. Schmidt PN, Roug S, Hansen EF, et al. Spectrum of microorganisms in infected walled-off pancreatic necrosis–impact on organ failure and mortality. Pancreatology 2014;14(6):444–9.
76. Tan C, Ling Z, Huang Y, et al. Dysbiosis of intestinal microbiota associated with inflammation involved in the progression of acute pancreatitis. Pancreas 2015;44(6):868–75.
77. Geng S, Yang L, Cheng F, et al. Gut microbiota are associated with psychological stress-induced defections in intestinal and blood-brain barriers. Front Microbiol 2019;10:3067.
78. Washio J, Sato T, Koseki T, et al. Hydrogen sulfide-producing bacteria in tongue biofilm and their relationship with oral malodour. J Med Microbiol 2005;54(9):889–95.
79. Forbes JD, Chen CY, Knox NC, et al. A comparative study of the gut microbiota in immune-mediated inflammatory diseases-does a common dysbiosis exist? Microbiome 2018;6(1):221.
80. Basic A, Blomqvist S, Carlen A, et al. Estimation of bacterial hydrogen sulfide production in vitro. J Oral Microbiol 2015;7(1):28166.
81. Lu H, Ren Z, Li A, et al. Tongue coating microbiome data distinguish patients with pancreatic head cancer from healthy controls. J Oral Microbiol 2019;11(1):1563409.
82. Salako NO, Philip L. Comparison of the use of the Halimeter and the Oral Chroma™ in the assessment of the ability of common cultivable oral anaerobic bacteria to produce malodorous volatile sulfur compounds from cysteine and methionine. Med Princ Pract 2011;20(1):75–9.
83. Zaric RZ, Jankovic S, Zaric M, et al. Antimicrobial treatment of morganella morganii invasive infections: Systematic review. Indian J Med Microbiol 2021;39(4):404–12.
84. Perler BK, Reinhart EM, Montgomery M, et al. Evaluation of the microbiome in men taking pre-exposure prophylaxis for HIV prevention. AIDS Behav 2021;25(7):2005–13.
85. Wei X, Tao J, Xiao S, et al. Xiexin Tang improves the symptom of type 2 diabetic rats by modulation of the gut microbiota. Sci Rep 2018;8(1):3685.
86. Tsai MC, Liu YY, Lin CC, et al. Gut microbiota dysbiosis in patients with biopsy-proven nonalcoholic fatty liver disease: A cross-sectional study in taiwan. Nutrients 2020;12(3):820.
87. Kort R, Schlösser J, Vazquez AR, et al. Model selection reveals the butyrate-producing gut bacterium Coprococcus eutactus as predictor for language development in 3-year-old rural Ugandan children. Front Microbiol 2021;12:681485.
88. Takada T, Kurakawa T, Tsuji H, et al. Fusicatenibacter saccharivorans gen. nov., sp. nov., isolated from human faeces. Int J Syst Evol Microbiol 2013;63(Pt 10):3691–6.
89. Takeshita K, Mizuno S, Mikami Y, et al. A single species of Clostridium subcluster XIVa decreased in ulcerative colitis patients. Inflamm Bowel Dis 2016;22(12):2802–10.
90. Liu X, Guo W, Cui S, et al. A comprehensive assessment of the safety of Blautia producta DSM 2950. Microorganisms 2021;9(5):908.
91. Ozato N, Saito S, Yamaguchi T, et al. Blautia genus associated with visceral fat accumulation in adults 20-76 years of age. NPJ Biofilms Microbiomes 2019;5(1):28.
92. Jenq RR, Taur Y, Devlin SM, et al. Intestinal Blautia is associated with reduced death from graft-versus-host disease. Biol Blood Marrow Transpl 2015;21(8):1373–83.
93. Meehan CJ, Beiko RG. A phylogenomic view of ecological specialization in the Lachnospiraceae, a family of digestive tract-associated bacteria. Genome Biol Evol 2014;6(3):703–13.
94. Takauji S, Konishi H, Fujiya M, et al. Polyphosphate, derived from Lactobacillus brevis, modulates the intestinal microbiome and attenuates acute pancreatitis. Dig Dis Sci 2021;66(11):3872–84.
95. Qin HL, Zheng JJ, Tong DN, et al. Effect of Lactobacillus plantarum enteral feeding on the gut permeability and septic complications in the patients with acute pancreatitis. Eur J Clin Nutr 2008;62(7):923–30.
96. Mangiante G, Colucci G, Canepari P, et al. Lactobacillus plantarum reduces infection of pancreatic necrosis in experimental acute pancreatitis. Dig Surg 2001;18(1):47–50.
97. Nagarchi A, Babu M, Saad S. Computed tomography severity index and serum CRP concentration for predicting the severity of acute pancreatitis. Int Surg J 2021;8(3):895.
98. Irshad A, Ahsan MF, Khan MA, et al. Correlation of C-reactive protein and computed tomography severity index in acute pancreatitis. Ann King Edward Med Univ 2021;27(2).
99. Mounzer R, Langmead CJ, Wu BU, et al. Comparison of existing clinical scoring systems to predict persistent organ failure in patients with acute pancreatitis. Gastroenterology. 2012;142(7):1476–82;quiz e1415-1476.
100. Langmead C, Lee PJ, Paragomi P, et al. A novel 5-cytokine panel outperforms conventional predictive markers of persistent organ failure in acute pancreatitis. Clin Transl Gastroenterol 2021;12(5):e00351.
