Community assembly modeling of microbial evolution within Barrett's esophagus and esophageal adenocarcinoma

Abstract

Mathematical modeling of somatic evolution, a process impacting both host cells and microbial communities in the human body, can capture important dynamics driving carcinogenesis. Here we considered models for esophageal adenocarcinoma (EAC), a cancer that has dramatically increased in incidence over the past few decades in Western populations, with high case fatality rates due to late-stage diagnoses. Despite advancements in genomic analyses of the precursor Barrett's esophagus (BE), prevention of late-stage EAC remains a significant clinical challenge. Previous microbiome studies in BE and EAC have focused on quantifying static microbial abundance differences rather than evolutionary dynamics. Using whole genome sequencing data from esophageal tissues, we first applied a robust bioinformatics pipeline to extract non-host DNA reads, mapped these putative reads to microbial taxa, and retained those taxa with high genomic coverage. When applying mathematical models of microbial evolution to sequential stages of progression to EAC, we observed evidence of neutral dynamics in community assembly within normal esophageal tissue and BE, but not EAC. In a case-control study of BE patients who progressed to EAC cancer outcomes (CO) versus those who had non-cancer outcomes (NCO) during follow-up (mean=10.5 years), we found that Helicobacter pylori deviated significantly from the neutral expectation in BE NCO, suggesting that factors related to H. pylori or H. pylori infection itself may influence EAC risk. Additionally, simulations incorporating selection recapitulated non-neutral behaviors observed in the datasets. Formally modeling dynamics during progression holds promise in clinical applications by offering a deeper understanding of microbial involvement in cancer development.

Keywords: Barrett’s esophagus; Esophageal adenocarcinoma; Helicobacter pylori; cancer microbiome; community assembly; mathematical modeling.

Figure 1.. Neutral and non-neutral community assembly and model dynamics in the context of the esophagus microbiome.

A. An overview of the model. B. Neutral model with mixed microbes in the community and an equal birth and death rates among all microbial taxa. C. Non-neutral model with potential domination of particular microbial taxa in the community enabled by different birth and death rates among microbial taxa.

Figure 2.. Alpha- and beta-diversity metrics across disease sets.

A. Microbial alpha-diversity represented by taxon richness (i.e., unique observed features) across disease stages. B. Microbial alpha-diversity represented by Shannon entropy across disease stages. C. Differences in community structure represented by the RPCA-PCoA between disease stages (PERMANOVA p=0.0001, pseudo-F=487). Shape corresponds to the disease state: square = Deshpande et al., 2018, circle = Paulson et al., 2022, diamond = Ross-Innes et al., 2015 & International Cancer Genome Consortium (ICGC) et al., 2020.

Figure 3.. Community assembly dynamics in progression to EAC.

Plots are ordered in sequence of disease state progression from healthy to EAC (see Supplementary Table S1 for patient/sample inclusion). The red curves represent the neutral model fit to data points representing unique taxa at the species level. The gray regions delineated by red dotted lines represent 95% bootstrap confidence intervals (obtained by resampling the hosts 100 times with replacement and refitting) and the $R^2$ value represents goodness of fit to the neutral model. The color of the data point represents phyla. Helicobacter pylori is colored pink with a box outline. Vertical line indicates mean relative abundance = 10⁻⁴. H. pylori is found with the following occurrence frequency in each group: Healthy 19.0476%, GERD 15.3846%, BE NCO 14.2857%, BE CO 5.0847%, EAC cohort 1 8.8235%, EAC cohort 2 0%.

Figure 4.. Simulation results for BE patients with non-cancer outcome (NCO).

A. Same data as shown for BE NCO panel in Figure 3, with neutral model fit for BE NCO patients. B. Neutral simulation with data from the BE NCO patients using the adjusted source pool conditions for H. pylori. All (‘A’) microbes have an equal birth and death rate of 1. C. Non-neutral simulation for BE NCO data with H. pylori adjusted source pool prevalence (H. pylori_SP =1e-5). For the H. pylori taxon, the birth and death rates are 1 and 0.5, respectively. For all other microbes, birth rates are drawn independently from a normal distribution (mean=1, standard deviation=0.1) and the death rate is set equal to 1. For A-C, the pink dot depicts data for H. pylori. For B-C, the source pool conditions of the H. pylori taxa are set to 1e-5 (see Results and Supplemental Figure S9 for details). SP = source pool; HP = H. pylori.

Figure 5.. ICGC EAC cohort 1 data with fits assuming neutral model at steady-state, neutral simulation, and non-neutral simulation.

A. Same data as shown in Figure 3, neutral model fit to data from EAC cohort 1 patients in ICGC dataset. B. Neutral model simulation with data from the EAC cohort 1 patients in ICGC dataset. All (‘A’) microbes have an equal birth and death rate of 1. C. Non-neutral model simulation with all microbes having variable birth rate drawn independently from a uniform distribution ranging from 1 to 6, and a death rate of 1. ICGC = International Cancer Genome Consortium.