Consistency across multi-omics layers 1 in a drug-perturbed gut microbial 2 community 3

Abstract

To investigate microbial community response to drug perturbations in a controlled system 103 across five omics layers, we combined 32 human gut microbiome representatives 104 (Tramontano et al, 2018) and exposed this community to three different non-antibiotic drugs 105 ( Figure 1A). The complete experiment was performed twice (run A and run B) as biological 106 replicates, starting from the initial community assembly step from single bacterial cultures. 107 More specifically, seven slow-growing species (inoculated on day 1) were combined with 25 108 fast-growing species (inoculated on day 3) on day 5 to form a synthetic community ( Figure 1A

122
After stabilisation, in each run the community perturbation was performed in duplicate during 123 exponential growth (i.e., five hours after passaging, as determined by optical density  dividing these values by the total protein intensity in each sample, as suggested previously 154 (Kleiner et al, 2017). 155 We compared relative species abundances between all pairs of omics methods except for 156 metabolomics, which by nature represents total metabolite measurements in the community 157 and does not allow to separate compounds by species. Based on correlation analysis, we 158 found the abundance estimates to be highly similar (minimum Spearman correlation coefficient 159 ρ = 0.78). Congruence was more pronounced for highly abundant species (Figure 2A). 160 Specifically, metagenomics and metatranscriptomics were the most similar of all pairwise 161 comparisons (ρ = 0.92). Further, 16S rRNA amplicon sequencing showed high similarity with 162 metagenomics for species with relative abundances higher than 0.001% (ρ = 0.89). However, 163 for several species with low relative abundances, 16S rRNA sequencing provided higher  methods, but only for species with relative abundance above 1% (ρ = 0.78 -0.84; 16 out of 172 6 29 species detected across all samples). This indicates that metaproteomics is less sensitive 173 than sequencing-based methodologies for species abundance estimation, as has also been 174 observed for in natura metaproteomics studies (Zhang & Figeys, 2019). Our results show 175 generally high consistency between omics data types in relative species abundance 176 estimations, and underline that metaproteomics can, in principle, provide robust species 177 abundance estimates, at least for synthetic microbial communities, albeit with lower sensitivity.   To systematically assess how much information on the functional level is captured by at least once). This is however expected as not all genes are expressed in any given condition.

214
Metaproteomics coverage was found to be much lower than metagenomics and  Second, only a subset of all metabolites present in the bacterial cell will be secreted outside 239 of the cell. Third, to calculate metabolic pathway coverage, we assumed that each pathway 240 consists of metabolites that are produced or consumed by metabolic enzymes annotated in 241 bacterial genomes, which is likely an overestimation of pathway sizes, since presence of an 242 enzyme in the genome does not necessarily imply that this enzyme is expressed, and that the 243 corresponding metabolite will be produced and measured extracellularly. Bacteroidota, Fusobacterium nucleatum was found to be less abundant in chlorpromazine-289 treated samples. In contrast, the other two drugs did not cause major shifts in relative 290 abundances: although ANCOM test identified significant changes of abundance of several 291 species, their relative abundance was not changing more than two-fold ( Figure 3B). In

302
Multi-omics measurements capture functional response of the 303 community to all three drugs 304 As compositional shifts do not provide information on the mechanisms of response of each 305 community member, we investigated these functional responses in more detail by performing 306 differential analysis of metatranscriptomic, metaproteomic and metabolomic datasets after a 307 normalization step wherein taxonomic abundance effects were reduced (see "Gene, transcript 308 and protein counting" in the Methods section). The highest number of differentially abundant 309 transcripts, proteins and metabolites were found in samples treated with chlorpromazine 310 (adjusted p-value < 0.001 and absolute fold change > 4 compared to control for 311 metatranscriptomics, adjusted p-value < 0.05 and absolute fold change > 1.5 for 312 metaproteomics and metabolomics; Figure 4A), which is in line with our findings that 313 chlorpromazine caused the largest disruption to bacterial community ( Figure 3B).

314
Transcriptional response to chlorpromazine is detected already after 15 minutes of treatment 315 across species belonging to different phyla, suggesting that, although Bacteroidota show the 316 strongest response, other species also adapt their gene expression.

317
In order to evaluate similarities between functional responses across omics data types, we   Bacteroides fragilis harbours multiple copies of RND pumps BmeABC (Ghotaslou et al, 2018). 418 Further, in addition to COG1538 (OprM homologues), also COG0841 containing homologues 419 of the MexB/AcrB/bmeB protein ( Figure 4E, also annotated in purple) was found to be enriched     In conclusion, we directly compared data from multiple omics methods and showed that they 510 agree on species abundance estimation of a defined and drug-perturbed microbial community 511 in vitro. Those methods that are able to detect functional information also correlate with each 512 other, albeit to a lower degree. We could also confirm expected time delays between 513 transcriptional and translational responses to perturbations, underlining that these methods 514 reveal biological insights that happen at different time scales. Although multi-omics analysis 515 of natural communities is hampered by their increasing complexity, combining multiple omics 516 measurements allows to measure the response of the community to perturbations across 517 molecular layers and provides information that is not achievable by any method alone.     . If the number of peptides was less than three, the protein was discarded.

715
To reduce taxonomic abundance effects in downstream analyses, taxon-specific scaling was 716 performed on metagenomics, metatranscriptomics and metaproteomics as described by Mantel test was performed to compare two different kinds of omics datasets and evaluate the 751 similarity between them. Abundance tables of each omics were transformed into distance 752 matrices using 1 -Spearman's correlation coefficient, and the matrices were compared using 753 the mantel function in the vegan package (version 2.5.5) with the default option. Sixty-one 754 samples that were common among all the omics datasets were used in this analysis.

755
Differential species abundance analysis 756 Differential analysis of species abundance across conditions was performed with ANCOM v.