Gut Microbial Shifts Indicate Melanoma Presence and Bacterial Interactions in a Murine Model

Abstract

Through a multitude of studies, the gut microbiota has been recognized as a significant influencer of both homeostasis and pathophysiology. Certain microbial taxa can even affect treatments such as cancer immunotherapies, including the immune checkpoint blockade. These taxa can impact such processes both individually as well as collectively through mechanisms from quorum sensing to metabolite production. Due to this overarching presence of the gut microbiota in many physiological processes distal to the GI tract, we hypothesized that mice bearing tumors at extraintestinal sites would display a distinct intestinal microbial signature from non-tumor-bearing mice, and that such a signature would involve taxa that collectively shift with tumor presence. Microbial OTUs were determined from 16S rRNA genes isolated from the fecal samples of C57BL/6 mice challenged with either B16-F10 melanoma cells or PBS control and analyzed using QIIME. Relative proportions of bacteria were determined for each mouse and, using machine-learning approaches, significantly altered taxa and co-occurrence patterns between tumor- and non-tumor-bearing mice were found. Mice with a tumor had elevated proportions of Ruminococcaceae, Peptococcaceae.g_rc4.4, and Christensenellaceae, as well as significant information gains and ReliefF weights for Bacteroidales.f__S24.7, Ruminococcaceae, Clostridiales, and Erysipelotrichaceae. Bacteroidales.f__S24.7, Ruminococcaceae, and Clostridiales were also implicated through shifting co-occurrences and PCA values. Using these seven taxa as a melanoma signature, a neural network reached an 80% tumor detection accuracy in a 10-fold stratified random sampling validation. These results indicated gut microbial proportions as a biosensor for tumor detection, and that shifting co-occurrences could be used to reveal relevant taxa.

A review of machine learning methods for cancer characterization from microbiome data

“…Unlike Decision Trees, Random Forest classifiers are highly popular when studying the association between cancer and the microbiome. This is supported by several published benchmarking experiments in which this model outperformed all other tested algorithms in tasks including identifying colorectal cancer 85 , melanomas in mice 87 , cancer subtypes 69 , and other host traits 41 . Random Forests have been applied to predicting the survival time of colorectal cancer patients from gene expression and microbiome taxonomic profiles 90 and identifying several tumor types such as epithelial ovarian cancer 44 , tonsillar squamous cell carcinoma 58 , lung adenocarcinoma 33 , colorectal cancer 28 , oral squamous cell carcinoma 98 , and in a multiclass classification setting 69 .…”

“…In this same publication, a boosting approach and a Random Forest demonstrated comparable performance 85 , while in ref. 87 a boosting model showed a decreased AUC but improved precision and recall over Random Forests. Boosting methods have also been proposed to predict tissue malignancy in breast cancer using bacterial taxonomic profiles from biopsies 30 and to identify several tumor subtypes from microbiome data 18 .…”

“…Feature extension can be done without increasing the computational load with the so-called kernel trick 86 . The most used kernel in tumor-associated microbiome analysis is the radial basis function (RBF) 85,87 , which has achieved good results in colorectal cancer prediction 48,71 . However, when kernels are used, the prediction for a sample is based on its kernel distance to the support vectors 88 .…”

“…SVMs have been used to predict colorectal patient survival time 89,90 , cancer prognosis, and drug responses 78 from microbiome and gene expression data, and to identify colorectal cancer patients using taxonomic profiles 71,85 . Furthermore, SVMs are frequently used as benchmarks when evaluating other methods 47,87,91 . The popularity of SVMs for cancer-related host trait prediction from the microbiome is due to their widespread availability in ML libraries and the good performance demonstrated by these models in multiple works.…”

“…85, L2 regularized Logistic Regression achieved a comparable AUC to Random Forest and Boosted Trees for colorectal cancer identification. However, this is seldom observed 49,87,91 ; as examples, Logistic Regression failed to improve colorectal polyp identification over a Multilayer Perceptron and Naïve Bayes classifier 92 , and colorectal cancer identification over a Multimodal Neural Network 47 . Because of this decreased performance, Logistic Regression has mainly been used for feature selection.…”