An In Vitro Medium for Modeling Gut Dysbiosis Associated with Cystic Fibrosis

Abstract

The gut physiology of pediatric and adult persons with cystic fibrosis (pwCF) is altered relative to healthy persons. The CF gut is characterized, in part, as having excess mucus, increased fat content, acidic pH, increased inflammation, increased antibiotic perturbation and the potential for increased oxygen availability. These physiological differences shift nutritional availability and the local environment for intestinal microbes, thus likely driving significant changes in microbial metabolism, colonization and competition with other microbes. The impact of any specific change in this physiological landscape is difficult to parse using human or animal studies. Thus, we have developed a novel culture medium representative of the CF gut environment, inclusive of all the aforementioned features. This medium, called CF-MiPro, maintains CF gut microbiome communities, while significantly shifting non-CF gut microbiome communities toward a CF-like microbial profile, characterized by low Bacteroidetes and high Proteobacteria abundance. This medium is able to maintain this culture composition for up to 5 days of passage. Additionally, microbial communities passaged in CF-MiPro produce significantly less immunomodulatory short chain fatty acids (SCFA), including propionate and butyrate, than communities passaged in MiPro, a culture medium representative of healthy gut physiology, confirming not only a shift in microbial composition but altered community function. Our results support the potential for this in vitro culture medium as a new tool for the study of gut dysbiosis in CF.

Keywords: colonoscopy; cystic fibrosis; dysbiosis; medium; stool.

Figure 1.. Impact of CF-MiPro on the diversity of cultured clinical samples from CF and nonCF subjects.

(A) Schematic of the experimental design. Stool and colonoscopy samples are grown in the indicated medium over 5 days under anoxic conditions, then subjected to analysis as follows: viable counting, amplicon sequencing and targeted analysis of SCFA. (B) Shannon Diversity Index (SDI) and (C) Chao-1 distance of CF (left) and nonCF (right) samples cultured in MiPro (purple), low-CF-MiPro (teal) and median-CF-MiPro (yellow) for five days. Each day, a 2% inoculum was introduced into fresh medium, with the passages every 24h. Samples were collected at day zero and daily though the end of day 5. A linear mixed effect model from the R package lme4 was used to test whether SDI or Chao-1 changed significantly with medium type. Patient was set as the random variable to control for multiple sampling. Both SDI and Chao-1 are significantly negatively correlated with low- and median-CF-MiPro for both genotypes, compared to MiPro. (D) Bray-Curtis beta diversity was calculated for each sample and displayed on a principal coordinate analysis (PCA) plot, faceted by genotype and colored by medium. The first two components account for 50.2% of total variance. (E) Significant differences in beta diversity were tested by PERMANOVA with metadata included in the model, and R-squared values for all metadata included in model, as well as potential interactions, are plotted with significance codes indicated (NS: non-significant, **: p<0.01, ***: p<0.001). “Residual” indicates variation unexplained by the model.

Figure 2.. Microbial relative abundance changes across media types.

Day of passage is graphed versus relative abundance of the indicated taxa for each sample, and a ribbon plot was used to visualize overall changes in microbial relative abundance at the (A) phylum level and (B) family level across all media conditions. The mean relative abundance is depicted by the line with 95% confidence interval shaded in grey. The legend indicates the taxonomic assignment for each panel. (C) Log2 fold change of taxa that were significantly (p-adj.<0.05) altered in nonCF samples passaged in median-CF-MiPro versus MiPro. Taxa were filtered to be present in at least 5% of all samples. Each dot represents a single genus and is color coded by association with CF gut dysbiosis (see main text for details). Significance was determined by DESeq2 using a non-continuous model of samples binned by media condition across days of passage 1 through 5. Patient was included as a design variable to control for multiple sampling.

Figure 3.. CFU/mL of nonCF and CF stool or colonoscopy samples passaged in CF-MiPro over five days.

NonCF (n=13) and CF (n=9) stool and colonoscopy samples were inoculated into each medium (MiPro, purple; low-CF-MiPro, teal; median-CF-MiPro, yellow) at a 2% final v/v ratio and passaged for 5 days, using this 2% inoculum into fresh medium daily. Each day, viable cell counts were quantified on sheep blood agar plates incubated at (A) 0% oxygen for obligate anaerobe and facultative anaerobe detection or (B) 21% oxygen for aerobic/facultative anaerobe detection. Total CFU/mL (log10(aerobic+anaerobic CFU/mL)) are displayed in (C). A linear mixed effect model from the R package lme4 was used to test whether CFU/mL within each genotype changed significantly with medium type and day of passage. Patient was set as the random variable to control for multiple sampling. CFU/mL are typically negatively correlated with CF-MiPro, in a dose-dependent manner, for the cultured CF and nonCF stool and colonoscopy samples, suggesting a decrease in richness. Overall, day of passage is positively correlated with CFU/mL for CF and nonCF cultured samples. To eliminate the contribution of Day of Passage, (D) anaerobic, (E) aerobic and (F) total CFU/mL from Day 5 are graphed by genotype, colored by medium. Another linear mixed effect model, with patient set as the random variable, was used to test whether Day 5 CFU/mL within each genotype changed significantly with medium. Under all oxygen tensions, CFU/mL from CF samples used as inoculum were not correlated with medium, while CFU/mL from nonCF samples used as inoculum were negatively correlated with median-CF-MiPro, but not low-CF-MiPro.

Figure 4.. Short-chain fatty acid (SCFA) production is decreased in CF-MiPro.

Filtered supernatants from Day 1–5 cultures were analyzed via GC/MS for quantification of the three most abundant SCFA: acetate, butyrate and propionate. The final concentrations were normalized to total anaerobic CFU/mL and graphed against Day of Passage for (A) CF and (B) nonCF samples. A linear mixed effect model from the R package lme4 was used to test whether each SCFA concentration within each genotype changed significantly with medium type and day of passage. Patient was set as the random variable to control for multiple sampling. While acetate concentrations were unaffected, butyrate and propionate concentrations were negatively correlated with CF-MiPro when CF samples were used as inoculum. In contrast, all three SCFA were negatively correlated with CF-MiPro when nonCF samples were used as inoculum. Day of passage inconsistently had a positive correlation with SCFA concentrations. To eliminate the contribution of Day of Passage, concentrations of each SCFA from Day 5 are plotted for (C) CF samples and (D) nonCF samples. Statistical analysis was performed using a similar linear mixed effect model as in (A) and (B). At Day 5, butyrate and propionate concentrations are negatively correlated with CF-MiPro across both genotypes of clinical samples.

Figure 5.. Short-chain fatty acid (SCFA) profiles are distinguished by medium in nonCF samples.

CFU-normalized concentrations of acetate, butyrate and propionate at Day 5 of passage were used to calculate Bray-Curtis beta diversity for (A) CF and (B) nonCF stool and colonoscopy samples. Ordination of each genotype was displayed in principal coordinate analysis (PCA) plots, colored by medium. For CF-derived samples, the first two components account for 72% of total variance. For nonCF-derived samples, the first two components account for 53% of total variance. Significant differences in beta diversity for (C) CF and (D) nonCF samples were tested by PERMANOVA with metadata included in the model, and R-squared values for Medium and Patient are plotted with respective p-values. “Residual” indicates variation unexplained by the model. SCFA profiles in nonCF samples are significantly distinguished by medium in a patient-independent manner, whereas SCFA profiles in CF samples are not distinguishable by medium.