Machine Learning of Pseudomonas aeruginosa transcriptomes identifies independently modulated sets of genes associated with known transcriptional regulators

Abstract

The transcriptional regulatory network (TRN) of Pseudomonas aeruginosa plays a critical role in coordinating numerous cellular processes. We extracted and quality controlled all publicly available RNA-sequencing datasets for P. aeruginosa to find 281 high-quality transcriptomes. We produced 83 new RNAseq data sets under critical conditions to generate a comprehensive compendium of 364 transcriptomes. We used this compendium to reconstruct the TRN of P. aeruginosa using independent component analysis (ICA). We identified 104 independently modulated sets of genes (called iModulons), among which 81 (78%) reflect the effects of known transcriptional regulators. We show that iModulons: 1) play an important role in defining the genomic boundaries of biosynthetic gene clusters (BGCs); 2) show increased expression of the BGCs and associated secretion systems in conditions that emulate cystic fibrosis (CF); 3) show the presence of a novel BGC named RiPP (bacteriocin producer) which might have a role in worsening CF outcomes; 4) exhibit the interplay of amino acid metabolism regulation and central metabolism across carbon sources, and 5) clustered according to their activity changes to define iron and sulfur stimulons. Finally, we compare the iModulons of P. aeruginosa with those of E. coli to observe conserved regulons across two gram negative species. This comprehensive TRN framework covers almost every aspect of the transcriptional regulatory machinery in P. aeruginosa, and thus could prove foundational for future research of its physiological functions.

Using genome-wide expression compendia to study microorganisms

“…Compendia can contribute to helping us gain a systems-level understanding of microbial biology. One major goal for systems biology is to model how information is encoded, specifically to reverse engineer the hierarchy of the transcriptomic regulatory network (TRN) [66] , [67] , [68] , [69] , [70] , [71] . Knowing the organization of a regulatory network allows us to control or optimize parts of the system, a necessary step for many biotechnological advances [72] , [73] , [74] , [75] .…”

“…A similar dimensionality reduction analysis was performed applying a sparse autoencoder to a yeast ( S. cerevisiae ) compendium, where Chen et al found latent variables represented pathways and other layers of biological abstractions such as a transcription factor complexes and signaling pathways [83] . In other studies, applying independent component analysis (ICA) to a compendium of transcriptome data revealed transcription modules [68] , [69] , [70] . Specifically, Poudel et al identified differentially active modules in S. aureus that varied based on the growth different media, which revealed metabolic regulators that respond to shifts in nutrients available [69] .…”

“…These sources of variance, which include noise generated by different experimental designs or different labs running the experiment, can confound the biological patterns that we are interested in detecting. Despite the presence of hundreds of technical sources of variability in compendia, previous compendia analyses have successfully extracted relevant biological patterns [70] , [80] , [81] , [83] . One hint for why these approaches work comes from a study by Lee et al .…”

Compendium-wide analysis of P. aeruginosa core and accessory genes reveal more nuanced transcriptional patterns

“…[17][18][19][20][21][23][24][25][26] However, how genetic differences translate into differences in the transcriptome and ultimately phenotypes is not well understood. The many transcription factors (TFs) and other transcriptional regulatory elements found in P. aeruginosa control the expression of many gene products that mediate various different traits 27,28 such as virulence [29][30][31] and these can vary across strains. For example, a study by Sana et al found a set of core genes that were differentially expressed in PAO1 compared to PA14.…”

Machine-learning fromPseudomonas putidaTranscriptomes Reveals Its Transcriptional Regulatory Network

“…However, we have demonstrated that iModulons recapitulate known regulatory mechanisms and can thus be considered as knowledge-based representations of the composition of the transcriptome. Indeed, recent studies successfully applied ICA for model bacteria (e.g., E. coli 13 , Bacillus subtilis 11 ) and even less-studied bacteria (e.g., Staphylococcus aureus 12 , Mycobacterium tuberculosis 19 , Sulfolobus acidocaldarius 19 , Pseudomonas aeruginosa 20 ) to unveil their genome-scale TRNs.…”