(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .



Spatial transcriptome-guided multi-scale framework connects P. aeruginosa metabolic states to oxidative stress biofilm microenvironment [1]

['Tracy J. Kuper', 'Department Of Chemical Engineering', 'University Of Virginia', 'Charlottesville', 'Virginia', 'United States Of America', 'Mohammad Mazharul Islam', 'Department Of Biomedical Engineering', 'Shayn M. Peirce-Cottler', 'Jason A. Papin']

Date: 2024-05

With the generation of spatially resolved transcriptomics of microbial biofilms, computational tools can be used to integrate this data to elucidate the multi-scale mechanisms controlling heterogeneous biofilm metabolism. This work presents a Multi-scale model of Metabolism In Cellular Systems (MiMICS) which is a computational framework that couples a genome-scale metabolic network reconstruction (GENRE) with Hybrid Automata Library (HAL), an existing agent-based model and reaction-diffusion model platform. A key feature of MiMICS is the ability to incorporate multiple -omics-guided metabolic models, which can represent unique metabolic states that yield different metabolic parameter values passed to the extracellular models. We used MiMICS to simulate Pseudomonas aeruginosa regulation of denitrification and oxidative stress metabolism in hypoxic and nitric oxide (NO) biofilm microenvironments. Integration of P. aeruginosa PA14 biofilm spatial transcriptomic data into a P. aeruginosa PA14 GENRE generated four PA14 metabolic model states that were input into MiMICS. Characteristic of aerobic, denitrification, and oxidative stress metabolism, the four metabolic model states predicted different oxygen, nitrate, and NO exchange fluxes that were passed as inputs to update the agent’s local metabolite concentrations in the extracellular reaction-diffusion model. Individual bacterial agents chose a PA14 metabolic model state based on a combination of stochastic rules, and agents sensing local oxygen and NO. Transcriptome-guided MiMICS predictions suggested microscale denitrification and oxidative stress metabolic heterogeneity emerged due to local variability in the NO biofilm microenvironment. MiMICS accurately predicted the biofilm’s spatial relationships between denitrification, oxidative stress, and central carbon metabolism. As simulated cells responded to extracellular NO, MiMICS revealed dynamics of cell populations heterogeneously upregulating reactions in the denitrification pathway, which may function to maintain NO levels within non-toxic ranges. We demonstrated that MiMICS is a valuable computational tool to incorporate multiple -omics-guided metabolic models to mechanistically map heterogeneous microbial metabolic states to the biofilm microenvironment.

Microbes secrete and respond to environmental metabolite signals, resulting in the spatial organization of heterogeneous physiological metabolic states within infectious microbial communities. Despite experimental advances, it is difficult to measure the connected, dynamic processes that control microbial community organization across multiple spatial and time scales. Thus, we developed an extendable multi-scale computational framework to simulate metabolic processes spanning intracellular to extracellular scales. We used this framework to simulate a Pseudomonas aeruginosa, a bacterium that causes lung infections, 3D biofilm community containing tens of thousands of single-cells. This is the first framework to integrate spatially resolved gene expression data. This data integration was advantageous to simulate cells regulating anaerobic, toxic metabolite byproduct secretion, and antioxidant metabolic states in response to a heterogeneous oxygen and toxic byproduct biofilm microenvironment. These cellular metabolic states were predicted to co-exist in biofilm spatial niches due to microscale variations in extracellular oxygen and toxic metabolite signals. Difficult to measure in experiments, this framework revealed multi-scale mechanisms underlying the emergent dynamics of each metabolic state, which may help to inform P. aeruginosa biofilm treatment strategies. We believe this multi-scale framework is valuable to integrate microbial data to uncover the multi-scale mechanisms regulating heterogenous metabolic processes in microbial communities.

Funding: All authors received financial support of U.S. federal funds from the National Institutes of Health (R01 AI154242 to JAP). T.J.K. was also supported by the National Institute of General Medical Sciences of the National Institute of Health under the award number T32GM136615. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Data Availability: Simulation inputs and source code files used in this work are available on Zenodo (DOI: 10.5281/zenodo.10493812 ). Simulation data generated in this work is available on Zenodo (DOI: 10.5281/zenodo.10494651 ). A detailed MiMICS User Guide and extendable code is available on GitHub at https://github.com/tracykuper/mimics .

Copyright: © 2024 Kuper et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

As an initial biological test case to demonstrate its utility, MiMICS was applied to simulate emergent metabolic heterogeneity within a 3D Pseudomonas aeruginosa biofilm observed by a recent spatial transcriptomic study [ 5 ]. P. aeruginosa is an opportunistic pathogen that can cause deadly biofilm infections in the lungs of patients with cystic fibrosis and COVID-19 [ 19 , 20 ]. The published spatial transcriptomic study revealed microscale spatial organization of aerobic, denitrification, and oxidative stress metabolic states within a P. aeruginosa PA14 biofilm [ 5 ]. As proposed by Dar and co-workers [ 5 ], genes related to denitrification metabolism, an anaerobic respiration process, were hypothesized to be upregulated in anoxic PA14 biofilm regions [ 5 , 21 , 22 ]. Likely secreted by denitrification cells, the cytotoxic denitrification intermediate nitric oxide (NO) was also hypothesized to upregulate oxidative stress genes in nearby PA14 biofilm cells [ 23 ]. However, the experiment lacked a quantitative and mechanistic mapping of the location of the cell and its metabolic state to the surrounding metabolic microenvironment[ 5 ]. Thus, herein, an established algorithm (RIPTiDe) [ 11 ] was used to integrate the published P. aeruginosa PA14 biofilm spatial transcriptomic dataset into a previously curated P. aeruginosa PA14 GENRE that generated four unique PA14 metabolic model states. The metabolic model states captured differences in aerobic and anaerobic denitrification metabolism, and revealed denitrification subpopulations that secreted the cytotoxic metabolite NO. This latter metabolic model state was crucial to predict a NO secretion rate that was passed to the extracellular reaction-diffusion model in MiMICS to generate a NO biofilm microenvironment that induced oxidative stress. Agents decided which metabolic model state to simulate their intracellular metabolic processes based on a combination of stochastic rules and metabolite sensing rules, the latter considering oxygen and NO levels in the agent’s local environment. MiMICS predicted microaerobic and variable NO microenvironments emerged within biofilm regions, resulting in microscale patches where cells heterogeneously used denitrification and oxidative stress metabolism. Due to cells sensing extracellular NO signals, MiMICS revealed the dynamics of cell populations heterogeneously regulating reactions in the denitrification pathway, which may function to maintain NO biofilm concentrations within non-toxic ranges. As demonstrated with this P. aeruginosa biofilm test case, we believe MiMICS is a promising computational tool that can use multiple -omics data-integrated metabolic models, and mechanistically simulate and map heterogeneous microbial metabolic states to the biofilm microenvironment.

In this work, we present an extendable multi-scale computational framework that couples multiple -omics data-integrated GENREs, an ABM, and metabolite reaction-diffusion PDEs. We refer to this framework as a Multi-scale model of Metabolism In Cellular Systems (MiMICS). MiMICS is an open-source Java- and Python-based framework. MiMICS is extendable to simulate in 2D and 3D, and to represent individual agents as a single cell or a population of cells. A key feature of MiMICS is the ability to incorporate multiple -omics data-integrated GENREs, which can represent unique metabolic states. As a result of the corresponding integrated -omics data, each metabolic model state may predict different parameter values that alter the extracellular environment, such as nutrient uptake or toxic byproduct secretion. MiMICS allows for the user to incorporate multiple GENREs integrated with -omics data measured at the single-cell or population-scale level. While MiMICS was designed to integrate spatially resolved transcriptomics data, MiMICS could be used to integrate global transcriptomics data that was measured in various metabolic conditions. MiMICS can execute biologically-relevant ABM rules for cellular agents to choose from the -omics data-integrated GENREs to simulate metabolism. Simple mechanistic rules were used for a cell to switch metabolic model states related to the cell’s extracellular metabolic environment, effectively representing the transcription of metabolic genes regulated by the metabolic environment. Future studies could directly couple MiMICS with a gene regulatory network, which predicts gene transcription regulated by the metabolic environment [ 18 ].

To simulate emergent spatiotemporal metabolic heterogeneity, computational frameworks have coupled GENREs with an agent-based model (ABM), which simulates individual cell behavior, and a reaction-diffusion model solved with partial differential equations (PDEs) that predicts extracellular metabolite concentrations. Previous 2D multi-scale frameworks, BacArena, MATNET, and COMETS, represent individual agents as a single cell or a population of cells, in which the simulation frameworks predicted emergent metabolic heterogeneity because each agent’s GENRE was constrained by local nutrient fluxes within a heterogenous nutrient environment [ 8 , 16 , 17 ]. Despite these efforts, few multi-scale frameworks are easily extendable for cells to dynamically adopt different -omics-integrated GENREs, which may improve predictions of metabolic processes controlled by gene regulation mechanisms. In addition, multi-scale frameworks have not incorporated GENREs integrated with spatial transcriptomic data, which can capture metabolic heterogeneity at single-cell spatial resolutions. For example, because the 3D multi-scale framework ACBM implemented a GENRE integrated with population-level transcriptomic data, this framework was not likely to capture potential biofilm metabolic heterogeneity measured at single-cell resolutions [ 15 ].

Although -omics data integration algorithms enable improved prediction accuracy of cellular metabolism, outputs from FBA simulations often represent the steady-state metabolism of a given cell type or species and thus do not capture metabolic heterogeneity across space and time. Dynamic FBA can be used to predict temporal changes in biomass and extracellular metabolite concentrations, but lacks consideration of metabolic spatial heterogeneity [ 13 ]. To reveal metabolic differences in space, previous efforts integrated spatial transcriptomics data from healthy and cancerous tissue regions into a human GENRE, but lacked predictions of dynamic cell-cell and cell-environment interactions to predict disease dynamics [ 14 ]. Despite these efforts, few computational frameworks have simulated -omics-integrated GENREs in both spatial and temporal dimensions [ 15 ], which is important to mechanistically predict pathogenesis and therapeutic outcomes in addition to other biological process.

One computational approach for simulating intracellular microbial metabolism uses genome-scale metabolic network reconstructions (GENREs). GENREs represent the complex, interconnected metabolic reaction network within a cell using a mathematical description of known gene-protein-reaction relationships and the stoichiometry of associated chemical transformations. GENREs can be interrogated with constraint-based flux-balance analysis (FBA) to simulate flux distributions associated with catabolic and anabolic processes [ 9 ]. Algorithms are being developed to integrate -omics data into a GENRE to constrain the intracellular metabolic solution space and generate biologically-relevant predictions of cellular metabolism in various metabolic environments [ 10 – 12 ].

Biofilms are three-dimensional, dynamic, heterogeneous microbial communities. Emergent biofilm metabolite gradients result in the spatial organization of bacteria with distinct metabolic functions, impacting the fate of the microbial community to expand, cause infection, and resist antibiotics [ 1 , 2 , 3 ]. Gaining a mechanistic understanding of biofilm metabolic spatial organization is limited by current experimental tools, making it difficult to measure the connected intracellular and extracellular metabolic processes across multiple temporal and spatial scales [ 4 ]. For example, at single-cell resolution, spatial transcriptomic experiments retain the spatial locations of bacterial metabolic states within biofilms [ 5 ], but they do not couple that information with high-resolution metabolomics measurements that can reveal mechanistic relationships between metabolic states and metabolite microenvironments [ 6 ]. To overcome these current experimental limitations, computational tools that bridge mechanisms across temporal and spatial scales can reveal underlying intracellular and extracellular mechanisms and prioritize experiments to test potential therapeutic strategies to treat infectious biofilms [ 7 , 8 ].

Results & discussion

MiMICS is an extendable computational framework executed in Python and Java to simulate metabolism in 2D and 3D microbial communities. MiMICS couples a genome-scale metabolic network reconstruction (GENRE) with the established platform Hybrid Automata Library (HAL) [24], which contains an agent-based model (ABM) and a continuum-scale reaction-diffusion model (Fig 1). To our knowledge this is the first multi-scale metabolic framework to interface with the COBRApy Python package, which is becoming increasingly common to simulate and integrate -omics data into a GENRE [11,25,26]. In addition, MiMICS is the first framework to couple an intracellular metabolic model with HAL, and simulate 3D microbial biofilms using HAL, which can be challenging due to the small microbial length scales that can cause instability in PDE solvers [24]. MiMICS offers the user the ability to input multiple -omics data-integrated metabolic models, which can represent distinct metabolic states and yield different metabolite uptake or secretion rates that are passed to the extracellular reaction-diffusion model. Individual agents decide which metabolic model state to execute based on mechanistic rules input by the user, such as agents’ sensing of their local metabolite concentrations.

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 1. Overview of the MiMICS computational framework. Individual bacterial agents were initialized at t = 0hrs in a 3D ABM world. Metabolites were initialized based on user-defined metabolite concentrations. For each five-minute simulation time step, the MiMICS framework was executed, consisting of a set of–omics data-integrated metabolic models, an ABM, and a metabolite reaction-diffusion model. MiMICS simulation outputs included extracellular metabolite concentrations, as well as each agent’s location, metabolic state, and intracellular metabolite fluxes at each time point. https://doi.org/10.1371/journal.pcbi.1012031.g001

In this study, an individual 2 μm x 2 μm x 2 μm agent represented a single-cell P. aeruginosa bacterium, which existed on a 230 μm x 230 μm x 40 μm three-dimensional grid, corresponding to experimental microscopy dimensions [5]. The continuum-scale reaction-diffusion model simulated oxygen, nitrate, nitric oxide, and glucose concentrations within an equivalently sized three-dimensional metabolite grid solved with partial-differential equations [24]. Agents were randomly initialized at t = 0hrs, and MiMICS simulated biofilm growth for ten hours, replicating the experimental biofilm growth period [5]. At each five-minute simulation time step, each model component of MiMICS was executed to update agent properties and metabolite concentrations (Fig 1). For each agent, the agent’s biomass and metabolite concentrations from the continuum-scale grid corresponding to the agent’s location were converted to metabolite uptake fluxes used to constrain the agent’s intracellular metabolic model (Fig 1). Constraint-based flux-balance analysis was used to optimize each agent’s metabolic model to predict a biomass growth rate, as well as metabolite secretion and uptake rates. The biomass growth rate was passed to the ABM to update an agent’s biomass (Fig 1). Each agent’s predicted metabolite secretion and uptake fluxes were passed to the reaction-diffusion model to update the metabolite concentrations at each agent’s location (Fig 1). In the ABM, bacteria agents performed cell division, moved via motility, and performed cell mechanical behaviors to prevent cell overlap (Fig 1). Simulation outputs such as agent locations, agent intracellular metabolic fluxes and metabolite concentrations were reported at desired simulation time steps.

Mechanistic incorporation of transcriptome-guided PA14 metabolic models into MiMICS MiMICS employs ABM rules for an agent to select a metabolic model state from a set of transcriptome-guided metabolic models input by the user. In this study, a combination of mechanistic ABM rules informed from literature and the spatial transcriptomic dataset were implemented into MiMICS (Table 1). P. aeruginosa has been observed to upregulate genes related to aerobic and denitrification metabolism in aerobic and deplete oxygen conditions, respectively [22]. In addition, in the presence of extracellular NO, P. aeruginosa increases expression of katA, encoding the antioxidant Catalase A [23]. Thus, to decide between the four PA14 metabolic model states, agents compared their local oxygen and NO concentrations to respective concentration thresholds, [O 2 ] t and [NO] t (Table 1). The parameter value for [NO] t was obtained from literature as the extracellular NO concentration that induced katA expression [23]. The parameter value for [O 2 ] t was fit to experimental outputs (S4 Fig). An oxic O 2 threshold, 0.21 mM, had the smallest MiMICS model error, which suggests complete oxygen depletion was not essential to induce denitrification metabolism. Indeed, previous studies observed P. aeruginosa used denitrification in microaerobic conditions (~0.05 mM oxygen) as a possible supplementary or competitive respiration strategy to aerobic respiration [28]. The parameterized [O 2 ] t value in the oxic range can be expected because the height (~10 μm) of the ten-hour biofilm was not large enough to predict significant oxygen depletion [33]. In the case that lower oxygen concentrations were present in the biofilm, other mechanisms that deplete oxygen near the biofilm, such as an oxygen boundary layer [34] or oxygen consumption by the planktonic phase above the biofilm may be present. These mechanisms serve as updates to simulate metabolite transport in upcoming MiMICS versions. Altogether, MiMICS suggests an oxygen transition point emerged within the PA14 biofilm that induced denitrification metabolism, but oxygen depletion mechanisms in regions outside of the biofilm may have been present. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Table 1. Mechanistic ABM rules for P. aeruginosa agents to choose a PA14 metabolic model state. P. aeruginosa agents compared their local oxygen and nitric oxide concentration to respective metabolite thresholds. A stochastic parameter, R n , determined the agent’s decision between the denitrification +/- NO secretion metabolic models. https://doi.org/10.1371/journal.pcbi.1012031.t001 A stochastic parameter, R n , generated only for agents in the denitrification-inducing low oxygen and low NO condition, was used for agents to select between the two denitrification metabolic model states, one with and one without predicted NO secretion flux (Table 1). The R n threshold was informed from experimental data, which suggested 6% of denitrification cells exhibited the norB-limiting gene expression profile associated with NO secretion (S2 Fig). This R n parameter suggests cells stochastically expressed nirS or norB, which encode nitrite reductase and NO reductase respectively. Indeed, stochastic expression of denitrification genes has been observed in other bacterial species [35]. In addition, experimental observations of P. aeruginosa biofilms after ten hours showed high expression levels of pilA (S1 Fig), which encodes type IV pili protein (PilA) that facilitates surface motility and shapes biofilm structure [5,36]. As pili synthesis reactions were not in the current PA14 GENRE [37], simple surface motility ABM rules were incorporated to recapitulate the PA14 biofilm structure and total cell count (S3 Fig).

Transcriptome-guided MiMICS predicted microscale metabolic heterogeneity and NO microenvironment in PA14 biofilm MiMICS was simulated with the mechanistic ABM rules controlling agent execution of a transcriptome-guided PA14 metabolic model state (Table 1) (referred to as transcriptome-guided MiMICS simulation). Qualitatively, in comparison to the experiment, the transcriptome-guided MiMICS simulation accurately predicted microscale, spatially confined biofilm niches in which cells heterogeneously upregulated denitrification and oxidative stress metabolic processes (Fig 3). In the simulation, these niches were located near the center of the biofilm colony, where microaerobic and variable NO concentrations were predicted (Fig 3). Thus, MiMICS simulations suggested that variable NO signal concentrations within microaerobic biofilm regions resulted in a heterogeneous population of cells using denitrification and oxidative stress metabolism co-existing within the same microscale niche. Qualitatively, experimental PA14 biofilms observed greater spatial dispersion of cells expressing denitrification and oxidative stress genes (Figs 3 and S5), suggesting other mechanisms may regulate denitrification gene expression, such as stochastic expression of the denitrification gene narG encoding nitrate reductase [35]. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 3. Transcriptome-guided MiMICS improved predictions of microscale metabolic heterogeneity and NO microenvironment in PA14 biofilm. Shown are representative 3D renderings of PA14 biofilms grown for ten-hours from the experiment, and transcriptome-free and transcriptome-guided MiMICS simulations. Cells are colored according to their metabolic state. Shown are 2D yz slices (x = 160 μm, 22 μm, 100 μm for experimental, transcriptome-guided MiMICS, and transcriptome-free MiMICS, respectively) of cell metabolic states, and predicted oxygen and NO concentrations. The x-values were chosen to compare similar biofilm colony structures across experimental and simulation conditions. Experimental data was reconstructed from Dar and co-workers. https://doi.org/10.1371/journal.pcbi.1012031.g003 To demonstrate the advantages of incorporating multiple transcriptome-guided metabolic models, MiMICS was simulated with the PA14 GENRE unconstrained by transcriptomic data (referred to as transcriptome-free MiMICS), which is the standard practice for current multi-scale metabolic frameworks [8,16,17]. Transcriptome-free MiMICS inaccurately predicted a homogeneous biofilm population with active flux through all intracellular denitrification metabolic reactions (Figs 3 and S5). In addition, transcriptome-free MiMICS did not predict extracellular NO in the biofilm microenvironment, resulting in agents lacking flux through the NO-induced oxidative stress reaction encoded by katA (Figs 3 and S5).

MiMICS captured spatial relationships of intracellular metabolism in PA14 biofilm To quantitatively evaluate the validity of transcriptome-guided MiMICS predictions in space, a bulk neighborhood spatial correlation analysis between genes was performed, similar to the analysis of experimental data by Dar and co-workers [5] (Fig 5B). Genes were compared which had both positive experimental expression values and simulated active flux values of reactions encoded by the respective gene. Shown in Fig 5B, MiMICS accurately predicted the spatial correlation for 33 gene pairs, and incorrectly predicted the spatial correlation for 22 gene pairs. Specifically, MiMICS accurately predicted that denitrification genes (i.e. narG, nirS, norB, nosZ) were positively correlated with one another, and positively correlated with the oxidative stress gene katA (Fig 5A and 5B). Conversely, MiMICS predicted napA, which encodes nitrate reductase, was positively correlated with the remaining denitrification genes narG, nirS, norB, and nosZ. However, this correlation was not observed in the experiment. Upon closer inspection, MiMICS predicted patches in the biofilm where neighboring cells utilized napA-, nirS-, norB-, and nosZ-encoded reactions (S6 Fig). In contrast, cells expressing napA in the experiment were more dispersed compared to more spatially confined biofilm regions of cells expressing nirS, norB, and nosZ (S6 Fig). MiMICS prediction discrepancies in the napA spatial correlations suggest alternate mechanisms exclusively modulate napA expression, such as extracellular phenazine secretion [39]. Indeed, phenazine synthesis genes phzE1 and phzM were expressed in the ten-hour PA14 biofilm (Fig 4). In addition, MiMICS accurately predicted denitrification and oxidative stress genes (i.e. narG, nirS, norB, nosZ, katA) were anticorrelated with the carbon metabolism genes gltB and sucC (Fig 5A and 5B). One aspect MiMICS did not accurately predict was the spatial correlation of atpA, encoding ATP synthase, with all other genes (Fig 5B), which motivates improvements in future transcriptome-guided MiMICS simulations. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 5. MiMICS captured spatial relationships of intracellular metabolism in PA14 biofilm. (A) Representative xy projections of PA14 biofilms from experiments and transcriptome-guided MiMICS simulations. Cells plotted are located near the z = 0 μm surface. Colored cells have high expression of the gene listed (experiment) or high reaction flux encoded by the gene listed (simulation). Circled areas highlight regions of interest where there is an anticorrelation between TCA cycle metabolism with denitrification and oxidative stress metabolism. Scale bar represents 20 μm. (B) Neighborhood gene spatial correlation comparison between experiment and transcriptome-guided MiMICS simulation. Spatial correlation between gene pairs was assessed by a Pearson correlation, where +1 and -1 value correspond to a strong positive and strong negative spatial correlation, respectively. The experimental and simulation Pearson correlation values are plotted in the upper right and lower left triangle of each square, respectively. Simulation Pearson correlation values were determined from 50 simulation replicates. Experimental data was reconstructed from Dar and co-workers. https://doi.org/10.1371/journal.pcbi.1012031.g005

[END]
---
[1] Url: https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1012031

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.