Duodenal Fluid Microbiome Diversity and Pancreatic Cyst Status Among Patients Undergoing Pancreatic Surveillance

Abstract

Background and Aims

Pancreatic cancer is associated with alterations in the gut microbiome; whether these changes are a cause or consequence of the disease is not known. This study aimed to compare the gut microbiome of patients with precancerous pancreatic cysts to that of patients without cysts.

Methods

Case/control analysis of 492 patients undergoing pancreatic surveillance, 267 patients with pancreatic cyst(s) and 225 without. Duodenal fluid bacterial DNA collected during endoscopic ultrasound was amplified and sequenced (16S ribosomal RNA). Measures of alpha and beta diversity were stratified by pancreatic cyst status, adjusting for demographic and clinical factors.

Results

Duodenal fluid alpha diversity measures were significantly lower in patients with pancreatic cysts compared to those without, with no significant differences between patients with small vs large pancreatic cysts, or those with worrisome features. Multivariate analysis revealed that measures of alpha diversity (Shannon index, operational taxonomic units, Faith phylogenetic diversity) were independently associated with having a pancreatic cyst: (Shannon index; adjusted odds ratio/95% confidence interval; 0.59/0.45–0.76, P < 01), along with diabetes status (2.35/1.2–4.79, P < .01) and age ≥75 years (2.72/1.43–5.50, P < .002). Duodenal fluid beta diversity differed significantly by covariates, including older age, proton pump inhibitor use, diabetes, current smoking, and regular alcohol use. In patients without these cofactors, duodenal fluid beta diversity did not differ significantly by pancreatic cyst status.

Conclusion

Patients undergoing pancreas surveillance who have pancreatic cysts have lower duodenal fluid alpha diversity compared to those without cysts. Further study is needed to determine if gut microbiome profiles predict future cancer risk in patients undergoing pancreas surveillance.

Graphical abstract

Introduction

Pancreatic cancer is the third most common cause of cancer death in the United States. The 5-year survival is among the lowest (∼13%) of all cancers¹ but is significantly higher when patients are diagnosed with localized disease,² especially Stage I disease.³ Individuals considered high risk based on the familial/genetic risk who maintain pancreas surveillance and are diagnosed with pancreatic cancer can achieve long-term survival.4, 5, 6 Pancreatic cancer develops through well-characterized precursor lesions, including pancreatic intraepithelial neoplasia and cystic neoplasms, the most common being intraductal papillary mucinous neoplasms (IPMNs). Unlike pancreatic intraepithelial neoplasia, cystic precursor lesions are macroscopically detectable and can be monitored with pancreatic imaging. In patients with suspected IPMN, the presence of worrisome pancreatic imaging features is associated with an increased likelihood of having high-grade dysplasia and/or invasive cancer, with the number of worrisome features being associated with a higher risk⁷ (reviewed in the study by Ohtsuka et al⁸). The identification of pancreatic cysts with worrisome-type features or isolated solid lesions can be an opportunity to intervene prior to the development of invasive cancer. However, the likelihood of having pancreatic cancer when worrisome features are present is low,9, 10, 11 and in patients under surveillance, pancreatic cancer often arises independent of existing pancreatic cyst(s).^4,12, 13, 14

Beyond pancreatic imaging, predicting future pancreatic cancer risk is challenging. Genetic risk, family history of the disease and the number of affected blood relatives with pancreatic cancer,¹⁵ smoking, older age, and diabetes are well-known risk factors.¹⁶ Factors associated with increased risk of neoplastic progression among patients under pancreatic surveillance, include having a mutation in a pancreatic cancer susceptibility gene,¹⁷ the number of a family members with pancreatic cancer,¹⁸ and the number and size of pancreatic cysts.¹⁹

Multiple studies have found the gut microbiome is altered in patients with pancreatic and other cancers, with some studies reporting high-specificity signatures.^20,21 However, since patients with pancreatic cancer often have other comorbidities, particularly metabolic disorders that can affect the gut microbiome,²² it is not known to what extent such changes arise prior to the development of cancer and if such changes influence the likelihood of neoplastic progression. Recently, a cohort study, in which prospective cases and controls were matched by age, gender, and race and ethnicity, found certain bacteria and fungi were risk factors for pancreatic cancer development.²³ Microbiomes have also been described within pancreatic cancer tissues that are associated with patient outcome.²⁴ However, recent studies find that for most cancer types, the tumor microbiome signatures are confounded by contamination.^25,26 Although the normal pancreas contains few bacteria, pancreatic tumor microbiomes could arise from both ductal spread such as when there is ductal dilation to the ampulla,²⁷ as well as from systemic spread. IPMN cyst fluid microbiomes have been analyzed with conflicting findings.^28,29

Most profiles of the gut microbiome utilize saliva or stool, though there are major differences between the oral and fecal microbiome.³⁰ In a prior report, we profiled the microbiome of duodenal fluid; it can be collected easily when patients undergo upper endoscopy and given its proximity to the pancreas. In that study, we found that patients with pancreatic cancer had reduced microbial diversity in their duodenal fluid microbiome compared with control subjects and patients with pancreatic cysts.³¹ If local microbiome factors are important contributors to pancreatic neoplastic development, profiling duodenal fluid microbiomes and identifying factors that influence it could yield more accurate microbiomes measure of pancreatic cancer risk.³¹

We hypothesized that if features of the gut microbiome influence pancreatic cancer risk such changes would be evident prior to the development of pancreatic cancer. To address this question, we compared duodenal fluid microbiome profiles of patients undergoing pancreatic surveillance.

Materials and Methods

Study Subjects and Sample Collection

Duodenal fluid samples were aspirated from the duodenal lumen during their endoscopic ultrasound from 553 participants enrolled in the Cancer of the Pancreas Screening (CAPS) studies^4,19 at Johns Hopkins Hospital between 2007 and 2023, mostly for their familial/genetic risk based on CAPS guidelines,^32,33 though we also included patients enrolled who had sporadic pancreatic cysts (ie, suspected IPMN without pancreatic cancer, n = 59). Sufficient duodenal fluid bacterial DNA for polymerase chain reaction and sequencing was obtained from 492 patients. We compared patient characteristics of these 492 patients with the 61 patients whose bacterial DNA concentrations after DNA extraction were insufficient; the only significant association was that regular proton pump inhibitor (PPI) users were less likely to have low duodenal fluid bacterial DNA concentrations (Supplementary Table 1). And among all 553 patients, the only factor significantly associated with duodenal DNA concentrations was regular PPI use, Supplementary Figure 1). Patients were grouped according to their pancreatic cyst status, presence vs absence, cyst number, cyst size, and the presence vs absence of worrisome features (Supplementary Table 2). Patients with pancreatic cancer were excluded. Covariate status at the time of duodenal fluid collection was extracted from the CAPS Research Electronic Database Capture (REDCap) database and with additional data not originally collected in REDCap (such as PPI use and fatty liver status) obtained from medical records. Duodenal fluid samples were aliquoted and stored at −80 °C. In a prior study, we evaluated the risk of microbial contamination during duodenal fluid collection by examining sterilized water drawn from 4 sterilized echoendoscopes and found no detectable microbial DNA.³¹ Subjects consented at enrollment to participate in studies focused on understanding risk factors and biomarkers for pancreatic cancer development and outcomes.

DNA Extraction

Duodenal fluid samples were pretreated with lysozyme (Boston BioProducts, Milford, MA). DNA was extracted using DNeasy blood and tissue kits (QIAGEN, Hilden, Germany) and bacterial DNA was quantified via real-time quantitative polymerase chain reaction (Femto Bacterial and Fungal DNA Quantification kits; Zymo Research, Irvine, CA) and stored at −80 °C. Samples with ≤0.002 ng/μL of bacterial DNA were not used for 16S ribosomal RNA sequencing. Sterilized water was also put through DNA extraction to check for potential microbial contamination; no microbial DNA was detected.

Bacterial Ribosomal RNA Gene Sequencing

DNA samples were amplified using the QIAseq 16S/ITS panel (Qiagen) using 1 μL of extracted DNA. Polymerase chain reaction cycle number was adjusted depending on the bacterial DNA concentration. Libraries were quantified with the QIAseq Library Quant Assay Kit (Qiagen). 2 nM of library (final concentration 6–10 pM) was sequenced with Miseq reagent kit v3 (600 cycles) (Illumina, CA). Raw sequences were processed with QIIME2 2023.2. Raw sequence data underwent demultiplexing and quality filtering with the 12-demux plugin, followed by denoising using DADA2. All amplicon sequence variants were aligned with Mafft a phylogeny was constructed using Fasttree2. Diversity metrics were calculated with q2-diversity after rarefying samples to 20,000 sequences per sample, excluding those below this threshold. Each amplicon sequence variant was identified using a Naïve Bayes classifier trained on 16S ribosomal RNA gene sequences from the SILVA ribosomal (rRNA) database (release 138).

Statistical Analysis

The primary analysis was to compare measures of microbial diversity by pancreatic cyst status. We also performed subgroup analyses, including cyst vs no cyst comparisons within the familial/genetic risk group (risk group), and among patients with pancreatic cysts, a comparison of familial/genetic vs sporadic cysts. Sample size was not formally calculated but estimated based on a prior study that included patients with pancreatic cysts.³¹ For alpha diversity, which reflects the complexity of microbial communities within individual samples, the following three complementary and widely used metrics were calculated: the Shannon index, the number of observed operational taxonomic units, and Faith phylogenetic diversity (these metrics measure dominance/entropy, richness, and diversity, respectively).³⁴ These indices were compared using the Steel–Dwass test. Pearson chi-squared test was used for categorical data. Beta diversity, which assesses differences in overall microbial community composition between groups or sites, was measured using Bray-Curtis distance and unweighted UniFrac principal coordinate analysis with P values determined through pairwise Permutational Multivariate Analysis of Variance (PERMANOVA). Differences in genera abundance were adjusted for multiple comparisons using the Benjamini-Hochberg method. Linear discriminant analysis effect size was utilized to identify the most significant differences in taxa between the groups. Firmicutes/bacteroidetes (F/B) ratios were determined and compared by cyst status and covariates using Wilcoxon. Multivariate analysis for alpha diversity and F/B ratios was performed using the least squares method. A multivariate logistic regression model was used to determine factors associated with the presence of pancreatic cyst(s). A P value below 0.05 was deemed statistically significant. All analyses were performed using JMP software (version 17.0.0; SAS Institute, NC).

Results

The characteristics of the 492 patients that had duodenal microbiome profiling are provided in Table 1. Approximately half the patients had 1 or more pancreatic cysts at the time of the duodenal fluid collection. Patients who had pancreatic cysts were significantly older and more likely to have diabetes. Most of the cohort (88%) was patients with familial/genetic risk and most patients with cysts and small, subcentimeter cysts typical of patients with familial/genetic risk. None of the patients had taken an oral antibiotic within the last 3 months.

Table 1.

Patient Characteristics by Pancreatic Cyst Status

Characteristic	Pancreatic cyst (n = 267)	%	No cyst (n = 225)	%	P
Clinical information
Age mean ± SD, y	66.5 ± 8.5		59.2 ± 9.9		< .001
Sex
Male	121	45.3	88	39.1	.165
Race
White	240	89.9	202	89.8	926
Black or African-American	9	3.4	9	4.0
Asian	6	2.2	4	1.8
Other	8	3.0	5	2.2
Unknown	4	1.5	5	2.2
BMI mean ± SDa	27.1 ± 5.2		26.2 ± 5.3		.775
Smoking status
Current	7	2.6	5	2.2	.253
Former	85	31.8	57	25.3
Never	175	65.6	163	72.4
Alcohol
Usual	85	31.8	75	33.3	.085
Occasional	41	15.4	50	22.2
None	141	52.8	100	44.4
PPI usersa	77	26.6	59	29.0	.561
H2 blocker usersa	8	3.0	8	3.6	.713
Probioticsa	20	7.5	7	3.1	.035
Diabetesa	38	14.3	13	5.9	.002
Fatty liver	75	28.1	53	23.6	.253
Pancreatic cyst
Cyst location
Head	128	47.9
Body	88	33.0
Tail	51	19.1
Cyst size, mean ± SD, mm	9.0 ± 8.5
Cyst number mean ± SD	2.2 ± 1.9
Worrisome features	30	11.2
Reasons for CAPS enrollmentb
Familial/genetic risk	208	77.9	225	100.0
Sporadic cyst	59	22.1	NA

CAPS, Cancer of the Pancreas Screening; SD, standard deviation.

^aFour cases with unknown data were excluded from the analysis of these factors.

^bFamilial risk are patients who meet surveillance criteria; ≥2 FDR (first-degree relatives), or 1 FDR + ≥ 1 SDR (second-degree relative), or 1 FDR with young onset pancreatic cancer ≤50 years old. Genetic risk indicates patients with a germline BRCA2, BRCA1, ATM, PALB2, CDKN2A variant, Lynch syndrome gene variants, and those Peutz-Jegher syndromes.

Patients with pancreatic cysts had significantly lower measures values for all three measures of alpha diversity than those without cysts (Figure 1). When we subdivided patients according to their pancreatic cyst size or cyst number, there was no significant difference in alpha diversity among patients with larger cysts (≥1 cm) compared to those with smaller cysts, or among patients with 1 or two vs three or more pancreatic cysts (Figure 1A and B).

Figure 1.

Duodenal fluid microbiomes in patients with vs without pancreatic cyst. (A) Analysis of alpha diversity measures in the duodenal microbiomes with patients with no pancreatic cyst, those with subcentimeter pancreatic cyst(s) and those with pancreatic cysts >1 cm diameter. (B) Analysis of alpha diversity measures in the duodenal microbiomes of patients with no pancreatic cyst, those with 1 or 2 cysts, vs those with 3 or more pancreatic cysts (C).

One or more measures of duodenal fluid alpha diversity were also significantly associated with patient age, alcohol consumption status, diabetes status (presence vs absence), and regular PPI use; no significant differences were found by obesity status (body mass index [BMI] <30/≥30), smoking status (never/former vs current), or fatty liver (Supplementary Figure 3). In a multivariate model, pancreatic cyst status and smoking significantly associated with 1 or more measures of alpha diversity. Other factors such as gender, regular PPI use, and diabetes status did not reach statistical significance (Table 2).

Table 2.

Multivariate Analysis for Factors Contributing to Duodenal Fluid Alpha Diversity

	Estimate	SE	95% CI		P
			Lower	Upper
Observed OTUs
Cyst	−34.20	7.70	−49.21	−19.15	< .001
Sexa	14.69	7.70	−0.41	29.80	.056
Diabetes	−23.60	12.61	−48.39	1.18	.062
PPI	−9.70	8.42	−26.25	6.84	.250
Probiotics	8.65	16.62	−24.00	41.31	.603
BMI over 30	1.33	9.27	−16.89	19.55	.886
Age 75 and older	−20.09	12.04	−43.75	3.57	.096
Current smoker	−9.81	24.03	−57.05	37.42	.683
Shannon index
Cyst	−0.144	0.04	−0.225	−0.07	< .001
Sexa	0.017	0.04	−0.05	0.09	.640
Diabetes	−0.070	0.06	−0.19	0.46	.238
PPI	−0.058	0.04	−0.09	0.14	.135
Probiotics	0.065	0.08	−0.09	0.22	.399
BMI over 30	−0.005	0.04	−0.08	0.09	.909
Age 75 and older	−0.050	0.06	−0.16	0.08	.374
Current smoker	−0.233	0.11	−0.45	−0.01	.039
Faith PD
Cyst	−0.65	1.93	−4.43	3.14	.738
Sexa	0.03	1.94	−3.83	3.77	.988
Diabetes	−4.37	3.17	−10.61	1.87	.170
PPI	−3.86	2.12	−8.03	0.31	.069
Probiotics	−1.21	4.18	−9.43	7.01	.772
BMI over 30	−0.57	2.33	−4.02	5.16	.808
Age 75 and older	−3.48	3.03	−9.44	2.48	.251
Current smoker	−0.11	6.05	−12.00	11.78	.985

SE, standard error.

^aReference is female.

Multivariate logistic regression model was utilized to identify factors independently associated with having pancreatic cyst(s). Measures of alpha diversity (Shannon index [odds ratio/95% confidence interval: 0.59/0.45–0.76, P < .001]), along with older age (≥75; adjusted odds ratio [aOR]: 2.72/1.43–5.50, P = .002) and diabetes (aOR: 2.35/1.22–4.79, P = .01) were significant predictors for having 1 or more pancreatic cysts (Table 3). Similar results were found for other measures of alpha diversity [observed operational taxonomic units (aOR: 0.67, 0.56–0.79, P < .001) and Faith phylogenetic diversity (aOR: 0.76, 0.64–0.90, P = .001)]. Among patients with pancreatic cyst(s), multivariate analysis yielded no other significant factors associated with alpha diversity (Supplementary Table 4).

Table 3.

Multivariate Analysis of Factors Associated With Having a Pancreatic Cyst

Characteristic	Odds ratio	95% CI		P
		Lower	Higher
Sexa	0.86	0.59	1.26	.441
Diabetes	2.35	1.22	4.79	.010
Current smoker	1.05	0.31	3.77	.932
Age 75 and older	2.72	1.43	5.50	.002
Shannon index	0.59	0.45	0.76	< .001

Sexa	1.25	0.85	1.84	.262
Diabetes	2.15	1.11	4.41	.022
Current smoker	1.22	0.37	4.32	.734
Age 75 and older	2.65	1.38	5.38	.003
Observed OTUsb	0.67	0.56	0.79	< .001

Sexa	1.20	0.82	1.76	.341
Diabetes	2.19	1.14	4.45	.018
Current smoker	1.32	0.41	4.57	.643
Age 75 and older	2.72	1.44	5.50	.002
Faith PDc	0.76	0.64	0.91	< .001

^aReference is female.

^bObserved OTUs stratified into four groups by quartile. Reference is lowest Observed OTUs group.

^cFaith PD was stratified by quartile. Reference is the lowest Faith PD group.

We performed a subgroup analysis within the 433 familial/genetic risk cohort to assess robustness. Although baseline characteristics differed between the 2 groups—specifically with respect to BMI, cyst size, and the presence of worrisome features, diversity findings were largely similar (Supplementary Figure 2, Supplementary Table 3A-E). Similar associations between duodenal fluid alpha diversity and covariates (age, diabetes, alcohol, smoking, PPI use) were found (Supplementary Figure 2B-I). A minor difference was observed in the multivariable Shannon index model (Supplementary Table 3D). When analyses were restricted to the familial/genetic risk cohort, worrisome features were no longer statistically significant in the Shannon index model, although effect estimates were otherwise similar (Supplementary Table 3D). We also compared alpha diversity in patients with pancreatic cysts by cohort (familial/genetic vs sporadic); no significant differences were found (Supplementary Table 3F).

Among patients with pancreatic cyst(s), we compared alpha diversity among patients with vs without worrisome features, none of whom had pancreatic cancer. Shannon index (P = .092) and Faith (P = .10) alpha diversity measures trended lower in patients with worrisome features (Supplementary Figure 4). Patients with worrisome features were somewhat older than pancreatic cyst patients without worrisome features (70 ± 7.9 vs 66.2 ± 8.8 years, P < .013). A multivariate logistic regression model to identify factors independently associated with having worrisome features among patients with a pancreatic cyst did not identify any measures of alpha diversity that were significant predictors; the only clinical factor associated with worrisome features was the risk group; more patients in the sporadic pancreatic cyst group had worrisome features, which likely reflects the propensity of these patients to be referred for endoscopic ultrasound (Supplementary Table 5).

We compared another common measure of gut health, the F/B ratio by pancreatic cyst status but found no significant association (Supplementary Figure 5). We did find that regular PPI users had higher F/B ratios than those who did not use PPIs. Older patients (age ≥75 years) and those with diabetes trended toward higher F/B ratios by univariate and multivariate analysis (Supplementary Table 6).

Patients with pancreatic cyst(s) had significant differences in beta diversity compared to patients without pancreatic cysts (both P = .001) (Figure 2A). Similar differences were found in beta diversity if we restricted the pancreatic cyst vs no cyst analysis to patients with familial/genetic risk (P = .001; data not shown). Notably, older age, regular alcohol consumption, diabetes status, PPI use, fatty liver, risk group, and smoking status were each also significantly associated with duodenal fluid beta diversity (unweighted UniFrac), and several covariates were significant by the Bray-Curtis measure of beta diversity (BMI and risk group; Supplementary Table 7). To assess if pancreatic cyst status might be independently associated with beta diversity, we analyzed the subset of patients without other features associated with beta diversity. Among patients without other risk factors associated with altered beta diversity (ie, patients under age 75, no regular PPI use, no diabetes, no fatty liver, not current smokers or regular alcohol users), neither the Bray-Curtis beta diversity measure nor the UniFrac measure differed according to pancreatic cyst status (presence vs absence) (Figure 2B). Within the subgroup of patients with pancreatic cyst(s), we did find a significant difference in beta diversity among patients with worrisome features vs those without (Figure 2C).

Figure 2.

Duodenal fluid beta diversity by pancreatic cyst status. (A) Beta diversity among patients with vs without pancreatic cysts using an unweighted UniFrac plot and Bray-Curtis dissimilarity plot, with P values determined by pairwise PERMANOVA. (B) Beta diversity by pancreatic cyst status in patients without risk factors for altered beta diversity (ie, limited to patient’s age <75 years, nonsmokers, non-PPI users, no regular alcohol consumption, no diabetes, no fatty liver). (C) Beta diversity by worrisome feature status in patients without risk factors for altered beta diversity (ie, limited to patient’s age <75 years, nonsmokers, non-PPI users, no regular alcohol consumption, no diabetes, no fatty liver).

We also compared microbiome abundance, genera, and class level according to pancreatic cyst status and other factors. Patients with pancreatic cyst(s) did not have any significant differences in the relative abundance of duodenal fluid microbes at the genus level or class level when adjusted for multiple comparisons (Supplementary Figure 6A). Similarly, linear discriminant analysis effect size analysis (Supplementary Figure 6B) showed no significant differences between patients with vs without pancreatic cyst.

Discussion

Given that prior research has shown a decrease in gut microbial diversity in patients with pancreatic cancer compared to controls,^20,21,31 our main objective was to determine if similar changes arise prior to the development of pancreatic cancer. Indeed, we find a notable reduction in duodenal fluid alpha diversity in patients with pancreatic cyst(s) compared to those without pancreatic cysts. The decrease in alpha diversity in patients with pancreatic cysts was independent of other factors associated with alpha diversity in our cohort, including older age, diabetes, and current smoking. Prior studies have found that measures of alpha diversity are significantly lower in older age subjects,35, 36, 37, 38 in diabetic patients,³⁹ current smokers, and regular PPI users.^31,40, 41, 42 We did not find significant differences in the ratio of the 2 major gut phyla, F/B, by pancreatic cyst status. Consistent with prior data, we found regular PPI use to be significantly associated with a high F/B ratio.^27,43 Although we also found a significant difference in beta diversity among patients with pancreatic cysts compared to those without, when we limited this comparison to patients free from covariates that associate with beta diversity (which in our cohort were patient’s age <75 years, non-PPI users, no diabetes or fatty liver, nonsmokers, and nonalcohol consumers), we did not find a significant difference, although we did find lower duodenal fluid beta diversity in pancreatic cyst patients with worrisome features compared to those without worrisome features.

Among patients undergoing pancreas surveillance for their familial/genetic risk, those with pancreatic cysts are at increased risk of pancreatic cancer.¹⁹ Among patients with incidentally detected pancreatic cysts, the risk is also considered elevated compared to population risk,^44,45 though studies show the risk of progression is modest overall,⁴⁶ with patients at greater risk of other diseases.⁴⁷ Whether the reduction in gut microbial diversity in patients with pancreatic cysts contributes to pancreatic cancer risk directly or merely reflects broader changes in host physiology that are contributing to pancreatic cyst development is not known.

Several recent studies have identified gut microbiome signatures that perform reasonably well in distinguishing patients with pancreatic cancer from controls, including disease controls with a variety of diseases,^20,21 although these studies did not compare these signatures to those of patients with pancreatic cysts without cancer. Many factors have been shown to affect the gut microbiome, including a variety of diseases.^38,48

Although many gut microbiome disease associations have been reported, few studies to date have shown that gut microbiome profiles can predict future disease. Several bacteria associated with pancreatic cancer risk in the oral microbiome cohort study have been implicated in periodontal disease,²³ which has been shown to be a risk factor for cancer.⁴⁹ A recent study found gut microbiome scores improved risk prediction of future coronary artery disease, diabetes, and prostate cancer risk predictive capacity over conventional risk factors alone.⁵⁰ In another study, gut microbiome profiles improved the prediction for future liver disease over conventional risk factors alone.⁵¹ Finally, a gut age index derived from 16S data in 2 cohorts had good accuracy for predicting common chronic diseases.⁵² Further studies are needed to determine the extent to which gut microbiome profiles can distinguish patients with pancreatic cancer from those with pancreatic cysts and if gut microbiome profiles can be used to better predict future pancreatic cancer risk.

Duodenal fluid microbiomes have certain differences compared to stool and saliva, including lower bacterial counts, a higher abundance of Proteobacteria, and lower abundance of Bacteroidetes. Most,^53,54 but not all,⁵⁵ prior studies have found that the duodenal fluid microbiome is more similar to the oral microbiome than the fecal microbiome.⁵⁶

Limitations

Certain limitations of our study to recognize include that it was retrospective single center study. Our study was designed to evaluate the pancreatic cyst/duodenal fluid microbiome associations. Most of our study population was patients under surveillance for their familial/genetic risk. The incidence of pancreatic cancer in our cohort is low; we did not have sufficient power to assess whether microbiome signatures predicted neoplastic progression. We were also underpowered to examine microbiome associations with patient comorbidities. Other factors known to influence the microbiome, including diet and geographic variation, were not systematically analyzed. Furthermore, additional gastrointestinal conditions that may affect the duodenal microbiome, such as functional dyspepsia, small intestinal bacterial overgrowth, and gastroparesis,57, 58, 59 were not systematically assessed. Although these conditions are not routinely encountered in our surveillance cohort, residual confounding related to unmeasured gastrointestinal comorbidities cannot be entirely excluded and should be considered when interpreting the findings. Prospective studies in larger cohorts are needed to confirm our findings.

Conclusion

We find reductions in duodenal fluid microbial diversity in patients with pancreatic cysts in their pancreas. Further research is necessary to explore the association between duodenal microbiome profiles and the development and progression of pancreatic cancer.