Microbial multiomics approach in understanding blue-veined cheeses bitterness

Abstract Blue-veined cheeses are regularly described as bitter but links between microbial diversity, function, gene expression and bitterness have not been deciphered yet. In the present work, microbiological, biochemical, metabarcoding (amplicons sequencing), metagenomic (DNA-Seq) and metatranscriptomic (RNA-Seq) were combined to better understand the potential origins of excessive bitterness in semi-soft blue-veined PDO Bleu d’Auvergne cheeses by exploring microbial composition, function, expression, conjugated to physico-chemical analysis and sensorial analysis. Physicochemical analyses revealed significative cheese bitterness differences associated with multiples factors in milk and in several cheeses’ parameters. After 30 days of ripening, Low Bitter (LB) cheeses were characterized by a higher bacterial and yeasts richness and diversity while High Bitter (HB) cheeses were characterized by one most dominant filamentous fungus Penicillium roqueforti and one most dominant bacterial species Lactococcus lactis. It affected in return the use of resources presents in milk through preferential mobilization of metabolic pathways as fatty acid desaturation (HB) and fatty acid biosynthesis (LB). Increase of salt concentration modulated the microbial community which withstood with osmostress. Abundance, richness, diversity of bacteria, eukaryote especially yeasts, phages and mycovirus were differentially involved in resources utilization from milk proteins, regulating cheese bitterness.

ADAMOS Analysis - Modular Structure This directory contains modular Quarto (QMD) files organized by data dependencies to recreate all figures and tables from the scientific paper: Microbial multiomics approach in understanding blue-veined cheeses bitterness, Michel et al.

Directory Structure analysis/ ├── README.md # Markdown version ├── README.html # This file ├── 01_physicochemical_sensorial.qmd # Figure 1, Table S1 ├── 02_amplicon_taxonomy.qmd # Figure 2, Figure S1, Table S2 ├── 03_shotgun_metagenomics.qmd # Figure 3, Table S3 ├── 04_rnaseq_pathways.qmd # Figure 4, Figure S2 ├── 05_multiomics_integration.qmd # Figure 5, Figure S3, Figure S4, Table S4 └── output_html/ # Rendered HTML reports (01-05)

R/ ├── helper_functions.R # Shared helper functions └── cog_cat.r # COG category definitions

data/ ├── amplicon_phyloseq/ # 16S and ITS phyloseq objects ├── shotgun_data/ # Shotgun metagenomics data ├── rnaseq_data/ # RNA-seq data ├── metadata_sup/ # Supplementary metadata ├── ko00001.rda # KEGG reference data └── KEGG_modules.rda # KEGG modules data

Analysis Files Overview

01_physicochemical_sensorial.qmd Data Required: Physicochemical metadata only Generates:

Figure 1: Boxplots of milk composition, sensorial scores, and cheese composition with statistical tests (t-tests with Bonferroni correction) Supplementary Table S1: ANOVA and Fisher's exact test results

Key Analyses:

Milk composition comparison (HB vs LB) Sensorial scores across ripening times Cheese physicochemical properties

02_amplicon_taxonomy.qmd Data Required: Amplicon phyloseq objects (16S and ITS) Generates:

Figure 2: Taxonomy barplots at genus level

  Panel a: Bacterial composition (16S, >1% abundance)
  Panel b: Fungal composition (ITS, >1% abundance)

Supplementary Figure S1: Alpha diversity boxplots with Tukey HSD

  Bacterial and fungal Observed and Shannon indices

Supplementary Table S2: DESeq2 differential abundance results

Key Analyses:

Taxonomic composition across bitterness classes and ripening times Alpha diversity (Observed richness, Shannon index) Differential abundance analysis (HB vs LB at 30 days)

03_shotgun_metagenomics.qmd Data Required: Shotgun metagenomics data (bins, taxonomy, metadata) Generates:

Figure 3a: Coverage scatter plot (HB vs LB bins) Figure 3b: Shannon diversity boxplots by domain Figure 3c: Virus-microbe correlation heatmap Supplementary Table S3: BUSCO statistics (completeness ≥70%)

Key Analyses:

Bin abundance comparison (log2 fold change) Shannon diversity across domains (Bacteria, Eukaryota, Viruses) Pearson correlations between viruses and microbes (|r| > 0.7)

04_rnaseq_pathways.qmd Data Required: RNA-seq data (RPKM, maaslin results, eggnog annotations) Generates:

Figure 4: KEGG pathway enrichment dotplots

  Panel a: LB30-associated pathways
  Panel b: HB30-associated pathways

Supplementary Figure S2: COG category bar plots

Key Analyses:

KEGG pathway enrichment (clusterProfiler) COG functional category distribution Differential gene expression (from Maaslin3 results)

05_multiomics_integration.qmd Data Required: All data types (amplicons, shotgun, RNA-seq, physicochemical) Generates:

Figure 5: Mixomics multiblock sPLS-DA

  Arrow plot showing sample projections
  Loadings plots for each data block

Supplementary Figure S3: Correlation heatmap (sensorial/physico/shotgun) Supplementary Figure S4: Mixomics component 1 correlations Supplementary Table S4: Correlation table (sensorial/physico/KEGG)

Key Analyses:

Multiblock sPLS-DA integration Cross-omics correlations Feature selection and importance

How to Use

Prerequisites Install required R packages:

Core packages

install.packages(c("tidyverse", "vroom", "ggpubr", "rstatix", "reshape2", "DT", "knitr"))

Microbiome packages

if (!require("BiocManager", quietly = TRUE)) install.packages("BiocManager") BiocManager::install(c("phyloseq", "DESeq2", "clusterProfiler", "enrichplot"))

Additional packages

install.packages(c("vegan", "microViz", "mixOmics", "mclust", "FactoMineR", "factoextra", "ComplexHeatmap", "circlize", "ggrepel", "ggh4x"))

Running Individual Analyses Each QMD file can be rendered independently:

Render a single analysis

quarto render analysis/01_physicochemical_sensorial.qmd

Render all analyses

quarto render analysis/

Or in R/RStudio:

Render a single file

quarto::quarto_render("analysis/01_physicochemical_sensorial.qmd")

Render all files

lapply(list.files("analysis", pattern = "\.qmd$", full.names = TRUE), quarto::quarto_render)

Output Each QMD file generates a self-contained HTML report with:

Interactive figures Statistical test results Session information for reproducibility

HTML files are saved in the same directory as the QMD files.

Data Organization The modular structure groups analyses by data dependencies:

Physicochemical only → File 01 Amplicons only → File 02 Shotgun only → File 03 RNA-seq only → File 04 All data integrated → File 05

This organization:

Minimizes data loading overhead Allows independent execution Makes debugging easier Facilitates parallel processing

Helper Functions All shared functions are in R/helper_functions.R:

add_sup_data() - Add supplementary metadata to phyloseq objects fill_tax_fun() - Fill taxonomy tables hierarchically load_physicochemical_data() - Load physicochemical metadata load_kegg_data() - Load KEGG reference data load_amplicon_data() - Load and prepare 16S/ITS data load_shotgun_data() - Load shotgun metagenomics data load_rnaseq_data() - Load RNA-seq data

These functions ensure consistency across analyses and reduce code duplication.

Troubleshooting

Common Issues

Issue: "Cannot find file 'data/...'" Solution: Ensure you're running from the project root directory or adjust paths in helper functions

Issue: "Package 'X' not found" Solution: Install missing packages (see Prerequisites section)

Issue: "Object 'kegg_dt' not found" Solution: Ensure load_kegg_data() is called before using KEGG functions

Issue: Memory errors with large datasets Solution: Increase R memory limit: memory.limit(size = 16000) (Windows) or adjust system settings

Citation If you use this code, please cite: Michel, E., Theil, S., et al. (2024). Microbial multiomics approach in understanding blue-veined cheeses bitterness. [Journal details]

Contact For questions or issues:

Elisa Michel: elisa.michel@inrae.fr Sébastien Theil: sebastien.theil@inrae.fr

Version History

v1.0 (2024-10-15): Initial modular structure

  Created files 01-04
  Implemented helper functions
  Documented data dependencies

R, 4.4.3