P266 Potential oral microbial markers for differential diagnosis of Crohn’s disease and ulcerative colitis using machine learning model

Park, S.K.(1);Kim, H.(2);Kim, S.(2);Lee, C.W.(3);Jung, Y.(4);Choi, C.H.(5);Kang, S.B.(6);Kim, T.O.(7);Bang, K.B.(8);Chun, J.(9);Cha , J.M.(10);Im, J.P.(11);Ahn, K.S.(12);Park, D.I.(1)*;

(1)Kangbuk Samsung Hospital- Sungkyunkwan University School of Medicine, Internal Medicine and Inflammatory Bowel Disease Center, Seoul, Korea- Republic Of;(2)Soongsil University, Bioinformatics, Seoul, Korea- Republic Of;(3)Kangbuk Samsung Hospital, Medical Research Institute, Seoul, Korea- Republic Of;(4)Soon Chun Hyang University- College of Medicine, Internal Medicine, Cheonan, Korea- Republic Of;(5)Chung-Ang University- College of Medicine, Internal Medicine, Seoul, Korea- Republic Of;(6)Daejeon St. Mary's Hospital- The Catholic University- College of Medicine, Internal Medicine, Daejeon, Korea- Republic Of;(7)Haeundae Paik Hospital- Inje University College of Medicine, Internal Medicine, Busan, Korea- Republic Of;(8)Dankook University College of Medicine, Internal Medicine, Cheonan, Korea- Republic Of;(9)Gangnam Severance Hospital- Yonsei University College of Medicine, Internal Medicine, Seoul, Korea- Republic Of;(10)Kyung Hee University Hospital at Gang Dong- Kyung Hee University College of Medicine, Internal Medicine, Seoul, Korea- Republic Of;(11)Seoul National University- College of Medicine, Internal Medicine and Liver Research Institute, Seoul, Korea- Republic Of;(12)PDXen Biosystems Inc, Functional Genome Institute, Seoul, Korea- Republic Of;

Background

Although the gut microbiome dysbiosis have independently been shown to be associated with inflammatory bowel disease (IBD), less is known about the relationship between oral microbiota and IBD. This study aimed to elucidate unique microbiome patterns in saliva from patients with IBD and investigate potential oral microbial markers for differentiating Crohn’s disease (CD) and ulcerative colitis (UC).


Methods

A multicenter, prospective cohort study recruited patients with IBD (UC, n=175, CD, n=127) and unrelated healthy controls (HC, n=100) to examine microbiota within the oral microenvironments. We used 16S rRNA gene sequencing data as features in training machine learning models (sPLS-DA, Sparse Partial Least-Squares Discriminant Analysis) to classify CD and UC.

Results

The V3-V4 amplicon reads of the saliva 16S rRNA sequencing data were taxonomically classified to a total of 2839 taxa (2270 genera) using Kraken2 based on Silva 138.1 reference. The sequences that were not classifiable down to family level were removed, and the samples having sequence depth less than 30000 were also removed, resulting in 2616 taxa for 390 samples (UC, n=168, CD, n=124, HC, n=98).
The alpha diversity analysis revealed that the microbiome in IBD patients were significantly less rich than the healthy controls, while CD samples were slightly richer then UC samples (Figure 1, Observed, P = 0.01, Shannon index, p=0.02, Chao index, P=0.0001). An sPLS-DA model with 470 taxa as features was able to distinguish IBD vs control with high performance (AUROC=0.9774), while a separate sPLS-DA model with 130 features classified CD vs UC with an AUROC of 0.8755 (Figure 2,3).

Conclusion

Collectively, oral microbial profiles can serve as a diagnostic marker to discriminate patients with IBD from HC, and patients with CD from UC. As obtaining oral samples is relatively easier than obtaining stool or intestinal biopsies, an opportunity exists to perform oral microbiome-based studies in larger cohort sizes, preferentially in a longitudinal fashion.