Microbiome changes associated with acute and chronic pancreatitis: A systematic review

Abstract

Background:

Altered intestinal microbiota has been reported in pancreatic disorders, however, it remains unclear whether these changes alter the course of disease in patients with acute (AP) and chronic pancreatitis (CP), or whether these disease states alter the environment to enable pathogenic microbial composition changes to occur. We undertook a systematic review to characterize the gut microbiome in pancreatitis patients.

Methods:

MEDLINE and EMBASE were searched for studies on microbiota in pancreatitis published from 1/1/2000 to 5/6/2020. Animal studies, reviews, case reports, and non-English articles were excluded. A frequency analysis was performed for outcomes reported in ≥2 studies and studies were analyzed for risk of bias and quality of evidence.

Results:

22 papers met inclusion criteria; 15 included AP, 7 included CP. No studies were appropriately designed to assess whether alterations in the gut microbiome exacerbate pancreatitis or develop as a result of pancreatitis. We did identify several patterns of microbiome changes that are associated with pancreatitis. The gut microbiome demonstrated decreased alpha diversity in 3/3 AP studies and 3/3 CP studies. Beta diversity analysis revealed differences in bacterial community composition in the gut microbiome in 2/2 AP studies and 3/3 CP studies. Functionally, gut microbiome changes were associated with infectious pathways in AP and CP. Several studies suffered from high risk of bias and inadequate quality.

Conclusions:

Detecting differences in microbial composition associated with AP and CP may represent a diagnostic tool. Appropriately controlled longitudinal studies are needed to determine whether microbiome changes are causative or reactive in pancreatitis.

Introduction

The human gastrointestinal tract hosts over 1000 different microbial species. These microbiota along with their genomes and the surrounding environmental conditions are collectively known as the gut microbiome¹. In healthy states, the gut microbiome promotes the development of the mucosal immune system and is involved in digestion, storage, and secretion of vitamins and nutrients^2,3. Recent evidence reveals that a diversified microbiome also exists in human blood in healthy and diseased states⁴. Many inflammatory, neoplastic, and metabolic conditions (including pancreatic disorders) are associated with microbiome changes^5–7.

Acute and chronic pancreatitis (AP and CP, respectively) are inflammatory conditions that have been associated with changes in microbial composition^8,9. The etiology and clinical features of AP and CP are heterogeneous and contribute to difficulties in disease diagnosis, characterization, and management. Current diagnostic modalities are unable to determine which patients will develop significant disease sequelae and complications; hence, there remains a need to better define disease progression in AP and CP. The objective of this systematic review is to characterize the taxonomic and functional characteristics of the gut and blood microbiomes in patients with different stages of pancreatitis. Summarizing these data may clarify whether changes in the gut and blood microbiomes exacerbate pancreatitis or develop as a result of pancreatitis.

Profiling the microbiome changes associated with AP and CP has the potential to influence the development of non-invasive prognostic tests and novel interventions that may alter the natural history of these diseases. However, while the future of this field is promising, the studies reviewed here highlight that this field is still in its early stages of investigation. This will specifically be addressed by our thorough review of the limitations of the studies included in this review.

Methods

The study protocol was registered with the international prospective register of systematic reviews with health-related outcomes (PROSPERO, CRD42020157560).

Search strategy, inclusion and exclusion criteria:

A comprehensive search of MEDLINE and EMBASE for relevant studies reported from 1/1/2000 to 5/6/2020 was performed. Reference lists from all studies retrieved by the original search strategies and conference proceedings were also queried to identify relevant studies. Primary database searches were performed using Medical Subject Headings (MeSH) to identify terms related to pancreatitis and the microbiome; these terms were then incorporated into the search along with standard title/abstract/keyword searches. Animal studies, reviews, case reports, and articles that were not available in English were excluded. The MEDLINE strategy is presented in Supplementary Figure 1.

The time point of 1/1/2000 was chosen as it was not until ~2005 that the advent of next-generation high-throughput sequencing enabled economically feasible metagenomic studies; prior to this many studies relied on culture-based techniques or antiquated analyses via fluorescent in situ hybridization (FISH)¹⁰. Additional exclusion criteria included studies focused on children and adolescents (under 18 years of age), or those including only patients with pancreatic ductal adenocarcinoma (PDAC), Cystic Fibrosis, autoimmune pancreatitis, or graft pancreatitis, or studies evaluating only helicobacter pylori. These conditions were excluded as their systemic impacts could result in microbiome changes that would confound those causing/resulting from pancreatitis. Studies focusing exclusively on helicobacter pylori were excluded as this organism is not representative of the gut microbiome. Other confounding variables that can affect the microbiome (antibiotic use, proton pump inhibitor use, metabolic conditions (pancreatic exocrine insufficiency, diabetes, obesity), alcohol intake, and current nutrition regimen were not considered exclusion criteria and the consideration of appropriate controls for these variables was assessed when grading the quality of evidence (described below). The literature search and screening of titles/abstracts was performed by two independent reviewers, and the full text of any title or abstract deemed eligible was obtained for analysis.

Data extraction (primary outcomes):

Data was extracted from all studies deemed eligible for inclusion by two independent reviewers using a Microsoft Excel spreadsheet (2016 Edition; Microsoft Corp., Redmond, Washington, USA) and discrepancies were resolved by discussion amongst reviewers. Data related to: study characteristics, patient demographic and disease characteristics, sample/assay used for microbiome analysis, abundance of bacteria in sample, diversity analysis, functional analysis (i.e. Kyoto Encyclopedia of genes and genomes (KEGG) pathway analysis), relative abundance of individual taxa, abundance of endotoxin, intestinal permeability, and intestinal absorption was extracted. The study was defined as using adequate diagnostic criteria for AP if the included subjects demonstrated at least 2 of the following features: characteristic abdominal pain; biochemical evidence of pancreatitis (i.e. amylase or lipase elevated >3 times the upper limit of normal); and/or radiographic evidence of pancreatitis on cross-sectional imaging¹¹. The study was defined as using adequate diagnostic criteria to grade severity of AP if criteria for the Atlanta classification were presented (1992 classification or 2012 revision depending on publication date)^11,12. The study was defined as using adequate diagnostic criteria for CP if subjects had at least one of the following: pancreatic calcifications visible on CT scan or endoscopic ultrasonography; moderate-to-severe pancreatic ductal lesions on endoscopic retrograde or magnetic resonance pancreatography (Cambridge classification); and/or histological features typical of pancreatitis on an adequate surgical specimen of the pancreas¹³.

The study was defined as assessing the gut microbiome if the sample used for analysis was derived from the alimentary tract or from the pancreas or biliary tract. The study was defined as assessing the blood microbiome if the sample used for analysis was derived from peripheral blood. Assays used to assess the microbial community were defined as culture-dependent or culture-independent. Culture-independent techniques were all based on the use of the 16S rRNA gene as a marker sequence (no studies used whole genome shotgun sequencing) and were further defined as limited detection (i.e. FISH, denaturing gradient gel electrophoresis, or targeted sequencing) or as high-throughput (i.e. unbiased sequencing of 16S rRNA gene amplicons)¹⁰. Papers were defined as performing a formal microbiome analysis if they performed a diversity analysis; diversity analysis includes a formal assessment of α or β diversity. α diversity (diversity within a community) reflects species richness (total number of species within a community), species evenness (relative distribution of species within a community), or a combination of these two elements¹⁴. β diversity (diversity between communities) is characterized by the number of species shared between two communities¹⁴. Because there was significant heterogeneity in the types of diversity analyses performed in each study, only categorical information for each variable was extracted (i.e. α diversity is listed as either increased/decreased/or not significantly different and β diversity is listed as either significantly or not significantly different).

Data extraction (risk of bias/quality of evidence):

For all journal articles meeting inclusion criteria, the risk of bias and quality of evidence was graded by two independent reviewers. The Cochrane risk of bias tool (RoB 2) and the quality assessment tool for diagnostic accuracy studies (QUADAS-2) were used to assess risk of bias in randomized controlled trials (RCTs) and observational studies, respectively^15,16. The Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach and the Newcastle-Ottawa Scale (NOS) were used to assess the quality of evidence in RCTs and observational studies, respectively. The RoB 2, QUADAS-2, and NOS tools use multiple signaling questions to assess risk of bias/quality within different domains (ex. The RoB 2 tool has 3 signaling questions to assess risk of bias for the randomization process domain). The signaling questions were answered “yes”, “no”, or “unclear” and are phrased such that “yes” indicates low risk of bias. If all signaling questions for a domain were answered “yes” then the risk of bias for that domain was judged “low”. If multiple signaling questions for a domain were answered “no”, then the risk of bias for that domain was judged “high”. Alternative scenarios were judged as “unclear”. The GRADE approach gives an initial score of 4 points (high quality evidence) to all RCTs but points can be lost or gained with respect to risk of bias, indirectness, inconsistency, imprecision, or publication bias; a score of 3 points represents moderate quality evidence, 2 points represents low quality evidence, and 1 point represents very low quality evidence¹⁷. The RoB 2 results for each RCT and the summative GRADE score for all RCTs are presented in a tabular form. For the QUADAS-2 and NOS analyses, the scores from the individual components of each tool are presented in graphical/tabular format and were not contorted into a single number/overall score because of the well-known problems associated with summary scores^18,19.

Synthesis of results:

Due to the substantial heterogeneity of study designs and outcomes reported, a frequency analysis of the 22 studies meeting eligibility criteria was conducted. Prior to frequency analysis, studies were divided into 4 comparison groups based on disease phenotype: 1) AP vs. healthy controls (HC); 2) Mild vs. severe AP; 3) Severe AP vs. severe AP + intervention; 4) CP vs. HC. In the event a single study did not meet inclusion criteria for one of these groups, the results were recorded for completeness but were not used for frequency analysis. If a study met inclusion criteria for multiple groups, it was analyzed within each group. If fewer than 2 studies/group reported an outcome of interest, that outcome was excluded from the frequency analysis. A comparison group for AP vs. CP was not included as none of the studies included data for both of these diseases.

Results

Search result and study selection:

Following the search strategy, a total of 223 articles were initially retrieved (Figure 1). 44 articles were removed as they were duplicates found in both searches. 151 articles were excluded based on title/abstract/keyword screening (28 animal studies, 44 reviews/case reports, 11 foreign language articles, and 68 articles that were excluded based on population (ex. Cystic fibrosis) or because they did not assess the microbiome as an outcome). This yielded 28 papers to assess in full for eligibility. After reviewing the full-text of these 28 papers, 6 were excluded for the following reasons: 1 was an abstract for which the same data was subsequently published as full text (full text manuscript was included in analysis), 1 precluded extraction of data specifically for patients with pancreatitis, 1 did not specify whether patients had acute or chronic pancreatitis, 1 did not include methods used for microbiome analysis, and 2 specifically analyzed organisms associated with infected pancreatic necrosis (not the microbiome). Therefore 22 studies were included in the qualitative synthesis (2 RCTs and 20 observational studies), examining a total population of 1,395 subjects.

Figure 1:

Database search and selection of studies according to PRISMA guidelines

Study characteristics:

A summary of the 22 included studies is shown in Table 1. 15/22 (68%) studies recruited patients with AP, 7/22 (32%) recruited patients with CP. Excluding conference proceedings, 11/13 (87%) AP manuscripts used adequate diagnostic criteria for AP and 3/6 (50%) CP manuscripts used adequate diagnostic criteria for CP. The average age of patients in AP studies was 52 years (IQR 44–56) and 48% were male (IQR 45–52%). The average age of patients in CP studies was 51 years (IQR 44–58) and 73% were male (IQR 58–83%). The primary etiology of pancreatitis was gallstones in 9/15 AP studies and was unidentified in 6/7 CP studies. 5/15 (33%) AP studies analyzed the blood microbiome and the remaining 10/15 (67%) analyzed the gut microbiome (8 used fecal samples, 1 used saliva, and 1 used a duodenal aspirate). 7/7 (100%) CP studies analyzed the gut microbiome (4 used fecal samples, 2 used saliva, and 1 used pancreas tissue). Culture-independent/high-throughput sequencing assays were used in 8/15 (53%) AP studies and 4/7 (57%) CP studies. 8/15 (53%) AP studies and 5/7 (71%) of CP studies performed a formal microbiome analysis.

Table 1:

Characteristics of the included studies

Author, yearref #	Publication type	Study type	Patients	Controls	Sample	Assaya	16S rRNA primers	Bioinformatics	Formal microbiome analysisb	Microbial change (patient vs. control)h
Ammori, 200327	Journal article	Longitudinal	- n=19 w/mild AP - n=7 w/severe AP -avg. age: 60 -% male: 58	- n=10 (healthy) -avg. age: 41 -% male: 40	- Blood - DNA extracted by cell lysis with SDS, Nonidet P40, and tween	Culture independent/limited detection (DGGE)	Primers not reported	NA	No	No bacterial DNA was found in any patient or control samples
Li, 201324	Journal article	Observational	- n=11 w/mild AP - n=37 w/severe AP -avg. age: 49 -% male: 48	- n=7 (healthy) -avg. age: 47 -% male: 43	- Blood - DNA extraction using QIAmp Mini Kit	-Culture independent/ limited detection (DGGE + sequencing of DGGE bands) - Sequencing via ABI PRISM 3730	F968GC and R1401 (V6-V8 region)	- Sequences compared with Genbank database - Multiple sequence alignments performed using CLUSTAL W - Neighbor-joining tree generated using MEGA	No	- No bacterial DNA was found in healthy control samples - In patients, 73% of sequences were Proteobacteria (P), predominantly Enterobacteriales and Pseudomonadales (O) and 27% of sequences were Firmicutes (P), predominantly Bacillales (O)
Li, 20187	Journal article	Observational	- n=50 w/severe AP -avg. age: 43 -% male: 60	- n=12 (healthy) -avg. age: 28 -% male: 83	- Blood - DNA extraction using QIAmp Mini Kit	-Culture independent/ high throughput - Sequencing via Ion Torrent PGM	357F and 518R (V3 region)	- Sequences clustered to OTUs with CD-HIT - OTUs taxonomically classified using ribosomal database project (RDP)	- α diversity estimated by OTU numbers - LEfSe analysis of enriched taxa	↑Bacteroidetes (P), Bacteroidia and Clostridia (C) ↓Actinobacteria (P), Actinobacteriae, Flavobacteriia and Bacilli (C)
Ma, 201928	Conference proceedings	Prospective	- n=33 w/severe AP + esomeprazole intervention -demographics not reported	- n=33 w/severe AP (no intervention) -demographics not reported	- Duodenal aspirate - DNA extraction method not reported	-Culture independent/high throughput - Sequencing via Illumina	Primers not reported	Not reported	- α diversity estimated by OTU count, Chao 1, Shannon, and Simpson methods - β diversity by principal component analysis - LEfSe analysis of enriched taxa	↑ Negativicutes (C), Selenomonadales (O), Veillonella (G), Shigella (G)
Madaria, 200555	Journal article	Longitudinal	- n=22 w/mild AP - n=9 w/severe AP -avg. age: 57 -% male: 42	- n=0 (none)	- Blood - DNA extracted using QIAmp DNA Mini Kit	-Culture independent/ high throughput - Sequencing via ABI PRISM Big Dye Terminator v3.1Cycle Sequencing Kit and ABI PRISM 310	27F and 1492R (nearly full length 16S rRNA)	- Sequences compared to RDP, GenBank, EMBL	No	NA
O'Keefe, 201120	Journal article	Longitudinal	- n=3 w/severe AP (before and after fiber supplementation) -avg. age: 72 -% male: 0	- n=5 (healthy, age and sex matched) -avg. age: 72 -% male: 0	- Feces - DNA extraction method not reported	-Culture independent/ limited detection (targeted sequencing) - Microarray using the Human Intestinal Tract Chip (HITChip)	Uni331F and Uni797R (V3-V4 region)	Not reported	No	↑ Bacteroidetes (P) ↓ Firmicutes (P), Clostridium clusters IV and XIVa
Philips, 201921	Journal article	Observational	- n=7 w/unspecified AP -avg. age: 41 -% male: 100	- n=7 (healthy) -avg. age: 41 -% male: 100	- Feces - DNA extracted via modified QIAmp DNA Stool Mini Kit1	-Culture independent/high throughput - Sequencing via Illumina MiSeq	Primers not reported	- Sequences classified taxonomically according to GreenGenes Database	- No α or β diversity analysis - LEfSe analysis of enriched taxa	↑ Actinobacteria and Verrucomicrobia (P), Moraxellaceae (F), Acinetobacter, Collinsella and Coriobacterium (G) ↓ Bacteroidetes and Fusobacteria (P), Prevotella (G) ↔ Firmicutes and Proteobacteria (P)
Poudel, 201937	Conference proceedings	Observational	- n=1 w/unspecified AP -demographics not reported	- n=9 (PDAC)c -demographics not reported	- Bile - DNA extracted using PowerViral RNA/DNA Isolation Kit	-Culture independent/high throughput - Sequencing via Illumina	Primers not reported	- Analysis using QIIME and MICCA	- Methods not specified	↑ Firmicutes (P), Clostridium sensu stricto (S) ↓ Proteobacteria (P)
Qin, 200830	Journal article	Prospective	- n=38 w/AP +intervention (parenteral nutrition) - n=36 w/AP +intervention (ecoimmunonut rition)e -avg. age: 56 -% male: 32	- n=10 (healthy) -demographics not reported -demographics not reported	- Feces - DNA extracted using a Fast DNA kit with CLS-TC lysis solution	Culture independent/limited detection (DGGE)	ITSPS2 and PL2 (V2-V3 region)	NA	No	NA
Tan, 20159	Journal article	Observational	- n=32 w/mild AP -n=44 w/severe AP -avg. age: 48 -% male: 48	- n=32 (healthy) -avg. age: 48 -% male: 56	- Feces - DNA extracted using QIAmp DNA Stool Mini Kit	- Culture independent/ limited detection (DGGE and targeted sequencing) - Quantitative polymerase chain reaction for 10 prespecified organisms.	341F and 534R (V3 region)	- DGGE analysis and similarities among DGGE profiles by Quantity One 1-D	- α diversity estimated by species richness (# of bands on DGGE)	↑ Bacteroides, Enterococcus, Enterobacteriaceae (G) ↓ Bifidobacteria (G) ↔ Clostridium and Lactobacteria (G), Faecalibacterium prausnitzii (S)
Wang, 201329	Journal article	Prospective	- n=61 w/severe AP +intervention(enteral nutrition) - n=62 w/severe AP +intervention (ecoimmunonut rition)f -avg. age: 43 -% male: 52	- n=60 w/severe AP +intervention(parenteral nutrition) -avg. age: 42 -% male: 57	- Feces - no DNA extraction	Culture dependent	NA	NA	No	↑ Bacillus bifidus and Bacillus acidi lactici (S) ↓ Enterococci (G) and Escherichia coli (S)
Yu, 202026	Journal article	Observational	- n=40 w/mild AP - n= 20 w/severe AP -avg. age: 44 -% male: 50	- n=20 (healthy) -avg. age: 32 -% male: 55	- Feces - DNA extraction method not reported	-Culture independent/ high throughput - Sequencing via Illumina MiSeq	Primers not reported	Not reported	- α diversity estimated by Shannon, Simpson, Ace, and Chao indexes - β diversity estimated using Bray-Curtis distance algorithm - Samples grouped according to partial least squares discriminant analysis (PLS-DA)	↑ Anaerococcus and Escherichia/Shigella (G) and Escherichia coli (S) ↓ Bifidobacteria, Blautia, Faecalibacterium, Prevotella and Subdoligranulum (G), and Eubacterium rectale (S)
Zhang, 200125	Journal article	Observational	- n=13 w/unspecified AP -avg. age: 53 -% male: 45	- n=10 (healthy) -avg. age: 53 -% male: 54	- Blood - DNA extracted using Chelex, SDS, Nonidet p40, and tween	Culture independent/ limited detection (agarose gel electrophore sis)	F101a and R356b (V2 region)	NA	No	NA
Zhang, 201822	Journal article	Observational	- n=45 w/mild AP -avg. age: 53 -% male: 45	- n=44 (healthy) -avg. age: 54 -% male: 55	- Feces - DNA extracted via QIAmp Media MDx Kit	- Culture independent/ high throughput - Sequencing via Illumina MiSeq	338F and 518R (V3 region)	- Sequences compared to dataset (specifics not reported) - Analysis via QIIME	- α diversity estimated by OTUs, Chao 1 and Shannon indexes - β diversity estimated by weighted UniFrac with principal coordinate analysis	↑ Bacteroidetes and Proteobacteria (P) ↓ Actinobacteria and Firmicutes (P)
Zhu, 201923	Journal article	Observational	- n=41 w/mild AP - n=89 w/severe AP -avg. age: 52 -% male: 49	- n=35 (healthy) -avg. age: 51 -% male: 51	- Feces - DNA extracted using Qiagen AllPrep DNA/RNA kit	- Culture independent/ high throughput - Sequencing via Illumina MiSeq	515F and 806R (V4 region)	- Sequences clustered as OTUs by scripts of USEARCH and taxonomically classified by RDP against Silva database - Metagenomes imputed by PICRUSt.	- α diversity estimated by observed OTUs - β diversity based on Bray-Curtis distances and principal coordinate analysis - Benjamini - Hochberg FDR correction for multiple hypothesis testing.	↑ Proteobacteria (P) and Enterobacteriaceae (F) and Escherichia/Shigella and Enterococcus (G) ↓ Bacteroidetes (P) and Bifidobacteria, Blautia, Faecalibacterium, lachnospiraceae and Prevotella (G)
Ciocan, 201831	Journal article	Observational	- n=24 w/alcoholic CP -avg. age: 51 -% male: 88	- n=45 (alcoholic) -avg. age: 51 -% male: 91	- Feces - DNA extracted using guanidinium thiocyanate lysis buffer	- Culture independent/ high throughput - Sequencing via Illumina MiSeq	PCR1F_460 and PCR1R_460 (V3-V4 region)	- Sequence clustering into OTUs based on Greengenes database - Analysis via QIIME,	- α diversity estimated by Shannon’s index - β diversity estimated by weighted and unweighted UniFrac distances - LEfSe analysis of enriched taxa - Benjamini- Hochberg FDR correction for multiple hypothesis testing.	↑ Proteobacteria (P), Aquabacterium, Enterococcus, Klebsiella, Pseudomonas and Sphingomonas (G) ↓ Bacteroidetes and Fusobacteria (P), Anaerostipes, Bacteroides, Bilophila, Lactococcus, Paraprevotella, Roseburia and Sutterella (G)
Farrell, 201152	Journal article	Observational	- n=27 w/CP -avg. age: 58 -% male: 56	- n=28 w/PDAC, 28 (healthy) -avg. age: 65 -% male: 64	- Saliva - DNA extracted using UltraClean Microbial DNA Isolation Kit	-Culture independent/limited detection (targeted sequencing) - Quantitative real time PCR for targeted organisms	27F and 1541R (nearly full length 16S rRNA)	Not reported	No	↑ Streptococcus mitis (S) ↓ Granulicatella adiacens (S)
Hamada, 201838	Journal article	Observational	- n=8 w/CP -avg. age: 60 -% male: 88	- n=12 w/AIPg -avg. age: 67 -% male: 83	- Feces - DNA extracted using PowerFecal DNA isolation kit	-Culture independent/ high throughput - Sequencing via Illumina MiSeq,	V4-V5 region (specific primers not reported)	- OTUs subjected to database analysis using Greengene and Living Tree databases	- α diversity estimated by OTU count	↔ Phylum level ↑ Bacteroides ovatus, Clostridium lactatifermentans, Clostridium lavalense, Streptococcus australis and Streptococcus gordonii (S)
Jandhyala, 20178	Journal article	Observational	- n=30 w/CP -avg. age: 32 -% male: 73	- n=10 (healthy) -avg. age: 42 -% male: 70	- Feces - DNA extracted using Qiagen mini stool DNA isolation kit	- Culture independent/ high throughput - Sequencing via Illumina MiSeq	338F and 806R (V3-V4 region)	- Taxonomic assignment using the MG-RAST server within RDP	- α diversity estimated by Shannon index - β diversity by Whitaker index - Bonferroni correction was applied for multiple hypothesis testing.	↑ ratio (Firmicutes: Actinobacteria) (P) ↓ Bacteroidetes (P), Faecalibacterium prausnitzii and Ruminococcus bromii (S) ↔ Class, order, family levels, Faecalibacterium (G)
Swidsinski, 200533	Journal article	Observational	- n=9 w/CP -demographics not reported	- n=0 (none) -demographics not reported	- Pancreas - DNA extraction method not reported	Culture independent/limited detection (FISH)	NA	NA	No	NA
Wang, 201953	Conference proceedings	Observational	- n=10 w/CP -demographics not reported	- n=10 (PDAC)d -demographics not reported	- Saliva - DNA extraction method not reported	Culture independentc	Primers not reported	Not reported	- Specific metrics for α and β diversity not reported.	Not reported
Zhou, 201932	Journal article	Observational	- n=71 w/CP -avg. age: 44 -% male: 58	- n=69 (healthy) -avg. age: 47 -% male: 42	- Feces - DNA extracted using Fast DNA SPIN Extraction Kit	-Culture independent/high throughput - Sequencing via Illumina MiSeq	338F and 806R (V3-V4 region)	- OTU taxonomy analyzed using RDP classifier algorithm - Sequence analysis using QIIME, R, and Majorbio ISanger platforms	- α diversity estimated by observed OTUs, Shannon, ACE, and Chao 1 - β-diversity estimated by weighted and unweighted unifrac distances - LEfSe analysis of enriched taxa.	↑ Proteobacteria (P) and Escherichia/Shigella, Parabacteroides and Prevotella (G) ↓ Actinobacteria and Firmicutes (P) and Faecalibacterium and Subdoligranulum (G)

^aCulture-independent techniques were all based on the use of the 16S rRNA gene as a marker sequence and were further defined as limited detection (i.e. FISH, denaturing gradient gel electrophoresis (DGGE), or targeted sequencing) or as high-throughput (i.e. unbiased sequencing of all of the 16S rRNA DNA amplicons)

^bIncludes either α or β diversity analysis

^cSpecific assay not specified

^dPancreatic ductal adenocarcinoma

^eEnteral nutrition + Lactobacillus plantarum

^fEnteral nutrition + Bacillus subtilis and Enterococcus faecium

^gAutoimmune pancreatitis

^hK=kingdom, P=phylum, C=class, O=order, F=family, G=genus, S=species

Studies on patients with AP vs. HC

We identified 10 studies comparing the microbiome in subjects with AP vs. HC. 6/10 (60%) analyzed the gut microbiome and 4/10 (40%) analyzed the blood microbiome. The frequency analysis for key study outcomes is presented in Table 3.

Table 3:

Frequency analysis for key study outcomes presented in studies comparing subjects with AP to HC

Comparison group	AP vs. HC
Sample	Gut microbiome		Blood microbiome
–# of papers in comparison groupref #	69,20–23,30		47,24,25,27
Key study outcomes	Reported in ≥ 2 studies?a	Resultb	Reported in ≥ 2 studies?a	Resultb
α diversity	Yes, 3/6	Decreased in 3/3 studies	No
β diversity	Yes, 2/6	Significantly different in 2/2 studies	No
LEfSec	No		No
Quantity of genetic information	No		Yes, 4/4	Increased in 3/4 studies
Specific taxa differences (phylum)	Yes, 4/6	Significantly different in 4/4 studies	No
Specific taxa differences (family)	No		No
Specific taxa differences (genus)	Yes, 3/6	Significantly different in 3/3 studies	Yes, 2/4	Significantly different in 2/2 studies
Specific taxa differences (species)	Yes, 2/6	Significantly different in 2/2 studies	No
Endotoxin level	Yes, 3/6	Increased in 2/3 studies	No
Intestinal permeability	No		No

^aPresented as yes/no followed by ratio of # reporting/total # of papers in comparison group

^bQuantities reported in terms of AP relative to HC (i.e. decreased = decreased in AP relative to HC)

^cLinear discriminant analysis Effect Size

Of the studies analyzing the gut microbiome in patients with AP compared to HC, there was 100% agreement between studies for key outcome variables. Specifically, α diversity was found to be decreased in patients with AP compared to HC, and β diversity was significantly different between these populations. The relative abundance of organisms within certain phyla, genus, and species levels was also noted to be significantly different between patients with AP and HC; however, when considering specific organisms there was significant heterogeneity in the results. There was disagreement regarding the level of Actinobacteria and Bacteroidetes in patients with AP compared to HC^20–23. There was agreement for Firmicutes and Bifidobacteria (decreased in AP vs. HC) and Proteobacteria and Enterococcus (increased in AP vs. HC)^9,20,21,23. 2/3 (67%) studies analyzing the endotoxin level noted it was increased in patients with AP compared to HC^9,23. Intestinal permeability was only assessed in one study which reported an increase in patients with AP compared to HC²³.

Of the studies analyzing the blood microbiome in patients with AP compared to HC, there were fewer key outcome variables reported and greater heterogeneity between study results. A diversity analysis was only performed in one study and the findings were similar to the gut microbiome; the blood microbiome was characterized by decreased α diversity and significantly different β diversity in patients with AP compared to HC⁷. 3/4 (75%) studies in this group reported an increased abundance of circulating DNA in the peripheral blood of patients with AP compared to HC^7,24,25. While this finding was considered to be a consequence AP-induced intestinal permeability and translocation of bacteria from the gut, none of these studies compared intestinal permeability between the patients with AP and HC. There was insufficient data to report how the abundance levels of specific organisms in the blood microbiome changed in patients with AP compared to HC.

Taken together, the gut and blood microbiomes in patients with AP and HC are diverse. There is inconclusive evidence to determine whether intestinal permeability and bacterial translocation are responsible for these findings.

Studies on patients with mild vs. severe AP

We identified 6 studies comparing the microbiome in patients with mild vs. severe AP. 3/6 (50%) analyzed the gut microbiome and 3/6 (50%) analyzed the blood microbiome. The frequency analysis for key study outcomes is presented in Supplementary Table 1. Only one study reported α diversity, which they found was not significantly different between patients with mild and severe AP⁹. Overall, the results for β diversity and the relative abundance level of specific organisms (i.e. Bacteroides) in the gut microbiome were discordant^9,23,26. However, there was agreement that β diversity was significantly different in patients with severe AP and infectious complications compared to patients with mild AP or patients with severe AP without infectious complications^9,23. 2/6 (33%) studies analyzed intestinal permeability and reported increased permeability was associated with disease severity in AP^23,27.

Studies on patients with severe AP vs. patients with severe AP + intervention

We identified 4 studies comparing the gut microbiome in patients with severe AP compared to patients with severe AP who received an intervention (esomeprazole (RCT, sample was duodenal aspirate)²⁸, ecoimmunonutrition (enteral nutrition + enteric coated Bacillus subtilis/Enterococcus faecium capsules (RCT, sample was feces)²⁹ or enteral nutrition + Lactobacillus plantarum solution (RCT, sample was feces)³⁰), or progressive fiber supplementation (longitudinal, sample was feces))²⁰. The frequency analysis for key study outcomes is presented in Table 4. Only one of these studies performed a formal microbiome analysis and reported that the β diversity was significantly different in patients with severe AP who received esomeprazole compared to those who did not²⁸. All 4 of these studies compared the relative abundance of organisms at the genus level between patients with severe AP and those with severe AP who received an intervention. Patients with severe AP receiving ecoimunonutrition and/or fiber had increased levels of Bifidobacteria and decreased levels of Enterococcus^20,29,30. Patients with severe AP that received ecoimmunonutrition and/or fiber also had improved clinical outcomes, fewer septic complications, reduced plasma endotoxin/TNF-α/IL-6 (inflammatory mediators), increased IL-10 (an anti-inflammatory cytokine), increased levels of fecal short chain fatty acids (a surrogate marker for metabolism by the microbiome), and normalization of their gut permeability^20,29,30. Taken together, although there is inconclusive evidence regarding the effect of proton pump inhibitors (PPI), ecoimmunonutrition, and fiber on the diversity of the gut microbiome in patients with severe AP, there is some evidence to suggest fiber and ecoimmunonutrition are associated with restoration of a healthy microbiome.

Table 4:

Frequency analysis for key study outcomes presented in studies comparing patients with severe AP compared to patients with severe AP who received an intervention

Comparison group	Severe AP vs. severe AP + interventiona
Sample	Gut microbiome
# of papers in comparison groupref #	420,28–30
Key study outcomes	Reported in ≥ 2 studies?b	Resultc
α diversity	No
β diversity	No
LEfSed	No
Quantity of genetic information	No
Specific taxa differences (phylum)	Yes, 2/4	Significantly different in 2/2 studies
Specific taxa differences (family)	No
Specific taxa differences (genus)	Yes, 4/4	Significantly different in 4/4 studies
Specific taxa differences (species)	Yes, 2/4	Significantly different in 2/2 studies
Endotoxin level	No
Intestinal permeability	No

^aIntervention: esomeprazole (RCT, sample was duodenal aspirate)²⁸, ecoimmunonutrition (enteral nutrition + enteric coated Bacillus subtilis/Enterococcus faecium capsules (RCT, sample was feces)²⁹ or enteral nutrition + Lactobacillus plantarum solution (RCT, sample was feces)³⁰), or progressive fiber supplementation (longitudinal, sample was feces))²⁰

^bPresented as yes/no followed by ratio of # reporting/total # of papers in comparison group

^cQuantities reported in terms of severe AP relative to severe AP +intervention (i.e. decreased = decreased in severe AP relative to severe AP+intervention)

^dLinear discriminant analysis Effect Size

Studies on patients with CP vs. HC

We identified 3 studies comparing the gut microbiome in patients with CP compared to HC. 3/3 (100%) studies used high throughput sequencing techniques, performed a formal microbiome analysis, and reached similar findings with respect to α and β diversity. Specifically, α diversity was found to be decreased in patients with CP compared to HC and β diversity was significantly different between these populations^8,31,32. Linear discriminant analysis effect size (LEfSe) was performed in 2 studies to determine whether the relative abundance of operational taxonomic units, a loose proxy for species, could be used as a marker to discriminate between patients with CP and HC; both studies reported the abundance levels of specific organisms distinguished the gut microbiome of patients with CP from HC—albeit with discordant organisms reported between studies^31,32. Specific organisms that were repeatedly found at different levels in the gut microbiome of patients with CP include Bacteroidetes (decreased in CP vs. HC)^8,31, Proteobacteria (increased in CP vs. HC)^31,32, and Faecalibacterium (decreased in CP vs. HC)^8,32. Proteobacteria was also increased in the pancreatic duct of patients with calcific CP and in the gut microbiome of patients with AP compared to HC^22,23,33.

Another similarity between the studies in patients with AP vs. HC and those with CP vs. HC is that the endotoxin level was also found to be increased in patients with CP compared to HC^8,32. LPS synthesis and plasma endotoxin levels also correlated with the presence of diabetes in patients with CP⁸. There was a significant negative correlation between the relative abundance of Faecalibacterium prauznitzii with plasma endotoxin levels and a positive correlation between the relative abundance of Shigella with plasma endotoxin levels^8,32. These results suggest individual microbiome constituents may be associated with an infectious phenotype in patients with pancreatitis.

Metabolic comorbidities (pancreatic exocrine insufficiency and diabetes) are associated with the morbidity of chronic pancreatitis. Multiple studies identified several species that, when altered in abundance, correlate with these comorbidities. There was a significant negative correlation between the relative abundance of Faecalibacterium prauznitzii with glycemic control^8,32. Akkermansia, an organism that has been negatively correlated with other inflammatory conditions in humans, and Bifidobacterium, an abundant commensal and probiotic, negatively correlated with pancreatic exocrine insufficiency^8,32,34. These results suggest microbiome changes are associated with CP severity.

Risk of bias and quality of evidence

2 RCTs were included in this systematic review and the risk of bias and quality of evidence for these studies was assessed using RoB2 and GRADE, respectively. RoB2 assesses risk of bias arising from the randomization process, deviation from the intended interventions, outcome measurement, missing outcome data, and data reporting in randomized controlled trials; GRADE assesses quality across the following domains: risk of bias, consistency, directness, precision, and publication bias^15,17. The results of these assessments are presented in Figure 2A. Within the RoB2 analysis, there were some concerns for each RCT regarding the effect of adhering to the intervention. This was considered a critical error by both reviewers and therefore resulted in a point reduction on the GRADE analysis. These RCTs were further downgraded on GRADE as they did not utilize high throughput sequencing for their microbiome analyses. Overall, these RCTs were designated low quality based on GRADE.

Figure 2:

Risk of bias and quality of evidence analysis of included prospective randomized controlled trials (Cochrane RoB 2 and GRADE) (A) and graphical representation of frequency of observational studies with low/high/unclear risk of bias based on QUADAS-2 (B), low/high/unclear applicability based on QUADAS-2 (C), or adequate quality based on NOS (D).

^aIntervention was a standardized amount of enteral nutrition; authors failed to address whether there were failures in implementing the intervention (i.e. patient developed ileus requiring cessation of enteral feeds) or if there were any instances of non-adherence to assigned intervention regimen (i.e. tube feeds on hold for a procedure).

^b Initial quality of RCTs (4 – High) was downgraded by 1 point due to concern for serious risk of bias and 1 point due to indirectness (methods used for microbiome analysis are not as robust as currently available high throughput sequencing methods). No points were lost/gained with respect to inconsistency, imprecision, or publication bias.

The remaining observational studies (excluding conference proceedings) were evaluated using QUADAS-2 and the NOS. QUADAS-2 evaluates risk of bias arising from patient selection, choice of index text, choice of reference standard, and flow and timing of patient testing¹⁶. The NOS assesses quality based on three broad perspectives: the selection of study groups, the comparability of the groups, and the ascertainment of either the exposure or outcome of interest for case-control or cohort studies, respectively³⁵. The subitem scores for study are presented in Supplementary Tables 2 (QUADAS-2) and 3 (NOS). A graphical representation of the results is depicted in Figure 2(B–C).

As shown in Figure 2B, a high percentage of studies had high risk of bias for patient selection and reference standard (15/17 (88%) and 17/17 (100%), respectively). The main sources of bias for patient selection arose from failure to define exclusion criteria and failure to specify whether a random or consecutive sample of patients was enrolled in the study. It was expected that 100% of the studies would have high risk of bias for reference standard as there is currently no available reference standard for the microbiome in pancreatitis. 4/17 (24%) studies had high risk of bias for the index test as they failed to specify an analysis plan or threshold value for their results. Only 1/17 (6%) studies suffered high risk of bias related to patient flow and this was because they failed to include all recruited patients in the analysis. With regard to applicability (Figure 2C), 12/17 (71%) studies had low risk of bias for patient selection; the 5 remaining studies were all in patients with CP and were not considered applicable as they selected patients who were admitted to the hospital with an acute exacerbation of CP and/or failed to provide the diagnostic criteria used to identify subjects with CP.

A majority of the observational studies had adequate quality for the following items: case definition, control selection, control definition, and ascertainment of exposure. Only 7/17 (41%) studies had an adequate representation of cases; the remaining 10 studies failed to define inclusion/exclusion criteria or suffered from selection bias as described above. Only 6/17 (35%) and 2/17 (12%) studies adequately controlled for major confounding variables (recent/current antibiotic use) and additional variables (acid suppression, metabolic disorders (including pancreatic exocrine insufficiency), diet, and alcohol intake), respectively. Finally, only 2/17 (12%) studies adequately defined the non-response rate within their study. Although several domains had an overall low risk of bias and adequate quality, the reviewers were most concerned with the high risk of bias regarding patient selection, the low applicability for the CP studies, and inadequate quality with respect to controlling for confounding variables.

Discussion

Several reviews have been published regarding the role of the microbiome in pancreatic health and disease. This is the first systematic review that focuses exclusively on pancreatitis. The most recent systematic review performed by Memba et al. in 2017 focused primarily on pancreatic ductal adenocarcinoma as, at that time, only 5 studies related to pancreatitis met their inclusion criteria³⁶. The present systematic review includes 22 studies and qualitatively analyzes the evidence for microbial composition changes in patients with AP vs. HC, patients with mild vs. severe AP, patients with severe AP who received an intervention, and in patients with CP vs. HC. Overall, we found that AP and CP are associated with microbial composition changes and the relative abundance of certain organisms could potentially be used as a diagnostic marker to identify patients with these conditions.

The abundance of Firmicutes was consistently reported to be decreased in AP vs. HC. Interestingly, a small study by Poudel et al. reported that the level of Firmicutes was even lower in patients with PDAC vs. AP^20,22,37. Faecalibacterium prausnitzii, one of the most abundant commensal organisms in the gut, is a member of the Firmicutes phylum; additional studies that define the microbiome at the species level will be important to determine whether reductions in F. prausnitzii are responsible for reducing the Firmicutes phylum in AP and PDAC.

In CP, several studies revealed that greater microbial composition changes and reduced relative abundance of commensal organisms are associated with CP disease severity^8,32. Similar findings have been reported regardless of the etiology of CP^32,38. The lack of association between etiology of CP and microbial composition changes, yet positive correlation between disease severity and these changes, suggests that microbial composition changes may be a result of CP rather than a cause.

This review reveals that certain interventions (i.e. ecoimmunonutrition and/or fiber) are associated with clinical and functional improvements in patients with AP. Unfortunately, formal diversity analyses were not performed in these studies—precluding researchers from determining whether these benefits are associated with normalization of the gut microbiome or the other known effects of fiber/probiotics³⁹. While these results may show promise for novel therapeutic approaches, it is important to consider that several of these agents have been used in RCTs before with mixed success. For example, the use of Lactobacillus+Bifidobacterium+fiber was associated with higher mortality and higher incidences of organ failure in the probiotic prophylaxis in predicted severe acute pancreatitis RCT, whereas Lactobacillus plantarum + fiber was associated with reduced incidences of organ dysfunction and pancreatic sepsis in two other RCTs^40–42. Importantly, the effect of these interventions on the gut microbiome and intestinal permeability was not assessed in these RCTs and the results could therefore be attributed to other effects of these interventions^40–42. Finally, it is interesting that none of the trials in this review utilized strains of Bifidobacterium (commonly used in probiotics) as this was one of the organisms noted to be depleted in patients with AP vs. HC, and depletion of this organism also correlated with disease severity in patients with CP. More research defining the species-level changes of the microbiome are needed before probiotics and other interventions, such as narrow spectrum antibiotics, can be used to promote commensal microbial recovery in patients with pancreatitis⁴³.

While this is the most comprehensive systematic review on this topic to date, several limitations need to be addressed. The first major limitation is the between study heterogeneity regarding methods used to assess and analyze the microbiome. Related to methodology, high throughput sequencing techniques have proved superior to culture-dependent techniques in several scenarios (e.g., infections caused by bacteria with unusual growth requirements, specimens collected during antimicrobial treatment)⁴⁴. Although 12/22 (55%) studies in this review utilized high throughput sequencing techniques, there were still inconsistencies in the results of diversity analyses between papers. A possible explanation for this discrepancy is that studies amplified different regions of the 16S rRNA gene (see Table 1)—a factor which has been shown to impact sequencing results⁴⁵. Also related to methodology, only 13/22 (59%) studies included a formal diversity analysis in their evaluation of the microbiome. Although diversity metrics themselves lack the resolution to be used as meaningful diagnostic or prognostic markers, diversity analyses are important tools to broadly assess microbial community composition as they provide a less biased approach to characterizing the microbiome compared to culture-dependent or targeted limited detection methods⁴⁶.

There was significant heterogeneity in the outcomes reported between studies. Many studies reported taxonomic data at different classification levels and used next-generation sequencing to report data at the species level. The current literature proposes using lower taxonomic ranks such as family, genus, or species, for 16S rDNA sequencing analyses, and to use whole genome sequencing to report data at the species level⁴⁷. Microbiome data from these lower taxa are more compatible with downstream metabolome analyses and the metabolome represents a bridge between the intestinal microbiota and host.

Another outcome variable that was heterogeneously reported was intestinal permeability. Although intestinal barrier dysfunction and translocation of bacteria may alter the disease course in patients with pancreatitis and is important to consider in the context of microbial composition changes, it is also important to consider that other conditions that are associated with intestinal barrier dysfunction (inflammatory bowel disease) are rarely associated with the severe complications observed in acute necrotizing pancreatitis. Several other pathogenic mechanisms (microcirculatory dysfunction and autodigestion) are more likely to be associated with the development of complications in acute pancreatitis^48,49. As such, future studies associating microbial composition changes with disease prognosis should control for all of these pathogenic processes.

Additional limitations of the studies included in this review relate to the applicability and quality of evidence. 5/17 (29%) of studies analyzing the gut microbiome used non-fecal samples; although we used this classification for the purpose of this comprehensive review, the oral microbiome (saliva) is typically analyzed independently of the gut microbiome (feces/intestinal mucosa). 5/6 (83%) of the included CP studies selected patients who were admitted to the hospital with acute exacerbations of their disease; it is unclear whether the microbiome findings in these patients are applicable to patients with quiescent CP. Related to the quality of evidence, the majority of studies included in this review did not control for antibiotic use or other confounding variables (e.g., PPI use, metabolic disorders (including pancreatic exocrine insufficiency), diet, alcohol intake). It is well established that host factors ranging from genes to diet influence the gut microbiome^50,51. One study demonstrated that there was no difference in the gut microbiome of patients with CP vs. PDAC when the analysis controlled for these variables; although this study was limited by a small sample size (n=20), the results disagree with similar studies that did not control for these variables^52,53.

All of the aforementioned limitations highlight how the results from Prospective Evaluation of Chronic Pancreatitis for Epidemiologic and Translational Studies (PROCEED) will have a marked impact on this field. PROCEED is the first prospective, observational cohort study of pancreatitis in the United States; enrollment began in 2017 and it seeks to enroll >1500 subjects⁵⁴. Factors such as medication use/diet/diabetes/pancreatic exocrine insufficiency will be controlled for in PROCEED. The longitudinal design of this study will allow subjects to serve as their own controls rather than relying on a reference standard which is not currently available. Finally, long term surveillance of patients with AP, recurrent AP, and CP, will enable researchers to define how microbiome changes are associated with the progression of these diseases.

Conclusion:

Microbial composition changes are associated with acute and chronic pancreatitis. The currently available correlational evidence is not suitable to determine whether these changes can be used clinically as a diagnostic or prognostic marker for these conditions. Future studies of the microbiome in pancreatitis could change our understanding of the pathogenesis of this disease and our approach to therapy.

Supplementary Material

Figure S1

Supplementary Figure 1: Example of search strategy used for studies related to pancreatitis and the microbiome

Table 2.

Summary of findings

	Key findings	Limitations	Recommendations for future studies
Acute pancreatitis	• Decreased alpha diversity • Differences in beta diversity • Decreased Firmicutes and Bifidobacteria • Increased Proteobacteria and Enterococcus • Association with infectious pathways	• High risk of bias associated with patient selection and representativeness of cases • Failure to control for confounding variables that affect the microbiome (i.e. antibiotics). • No defined reference standard for a “healthy” microbiome • Heterogeneity in methods used by each study to assess and analyze the microbiome	• Utilize metagenomics rather than amplicon-based (16S rRNA) methods • Concordantly assess the oral, gut, and blood microbiomes and metabolomes in patients and controls • Perform longitudinal assessments of the microbiome to define how microbiome composition changes are associated with progression of disease (ex. Patients with AP who develop CP or patients with CP who develop CP + DM or PEI).
Chronic pancreatitis	• Decreased alpha diversity • Differences in beta diversity • Increased Proteobacteria • Decreased Bacteroidetes and Faecalibacterium • Association with infectious pathways • Association of microbial community changes with metabolic comorbidities (DM and PEI)

Table 5.

Frequency analysis for key study outcomes presented in studies comparing patients with CP to HC

Comparison group	CP vs. HC
Sample	Gut microbiome
# of papers in comparison groupref #	38,31,32
Key study outcomes	Reported in ≥ 2 studies?a	Resultb
α diversity	Yes, 3/3	Decreased in 3/3 studies
β diversity	Yes, 3/3	Significantly different in 3/3 studies
LEfSec	Yes, 2/3	Significantly different in 2/2 studies
Quantity of genetic information	No
Specific taxa differences (phylum)	Yes, 3/3	Significantly different in 3/3 studies
Specific taxa differences (family)	Yes, 2/3	Significantly different in 1/2 studies
Specific taxa differences (genus)	Yes, 3/3	Significantly different in 3/3 studies
Specific taxa differences (species)	No
Endotoxin level	Yes, 2/3	Increased in 2/2 studies
Intestinal permeability	No

^aPresented as yes/no followed by ratio of # reporting/total # of papers in comparison group

^bQuantities reported in terms of CP relative to HC (i.e. decreased = decreased in CP relative to HC)

^cLinear discriminant analysis Effect Size