Critical need to improve soil health and low-input bioenergy crop production. It is widely recognized that poor land management practices have resulted in extensive soil erosion, degradation, and loss of soil organics. Bioenergy crops and beneficial microbiomes have the potential to both restore soil health and at the same time support the growing bioeconomy. Unfortunately, there are basic science gaps that we must address to achieve this vision. Specifically, we don’t know the composition of soil organics and how they contribute to soil and plant health, nor do we understand how soil organics are produced by plants and microbes. To understand these molecular processes impacting soil organics, we use fabricated ecosystems, exometabolomics, robotic automation, and advanced computing (AI/ML). The knowledge gained will enable predictive control of soil systems using targeted interventions.

Index

• Exometabolomics and Environmental Metabolomics
• Ecosystem Fabrication and Autonomous Experimentation
• Artificial Intelligence and Informatics

Exometabolomics and Environmental Metabolomics

Exometabolomic approach: In an exometabolomic experiment, microbes are grown in a medium containing many metabolites. Comparison of the microbially conditioned media vs. control media reveals which metabolite were used and produced by the microbes. We use this type of analysis for a wide-range of systems especially soil and sediment microbes. Since a large fraction of the carbon fixed by plants is released as root ‘exudates’ to recruit and retain beneficial microbes, we use the same approach to characterize root exudates. By growing rhizosphere bacteria in the root exudates we predict metabolic exchanges between microbes and with the plant host (among other things). Understanding this process is critical to low-input bioenergy production and management of soil health.

Microbial substrate preferences dynamically link soil microbes to soil metabolites: Historically, microbes have been studied in isolation for their growth on single substrates. However, in natural environments microbes have access to a ‘buffet’ of diverse substrates. Exometabolomics enables us to discover which metabolites are preferred by specific microbes and we find that this is highly specific, sometimes even distinguishing between different bacterial strains. Building on an earlier study by our lab (Baran et al 2015) that defined the substrate preferences of biological crust soil microbes we found that these preferences explained the correlations between bacterial and metabolite abundance after wetting the soil (Swenson 2018) and in collaboration with other groups found that this analysis could be improved using AI/ML (Morton 2020). This approach is also used to improve our understanding how groundwater and sediment organics support important microbial activities in the subsurface. Environmental microbes perform a vast array of metabolic processes. As part of the ENIGMA consortium, we are understanding how subsurface organics support microbial communities that attenuate the movement of heavy metals and manage nitrate at a contaminated field site.

Web of Microbes is both repository for our exometabolomics data experiments and tool for microbiome design. It capture assertions of metabolites compositions in environments and changes to metabolite abundances by organisms in the environment. For more information, visit the resources tabor visit webofmicrobes.org

Ecosystem Fabrication and Autonomous Experimentation

Workflow for developing fabricated ecosystems (EcoFABs) to rapidly improve our understanding of important molecular interactions within important soil environments like the rhizosphere. These systems address key challenges within microbiome research. First, they enable replication across labs (Novak and Andeer et al, 2024), which is a key gap in microbiome research. Second, because they are standardized they can produce AI ready data to accelerate scientific discovery.

Spanning scales and complexity: In collaboration with many other labs, we have constructed fabricated ecosystems that span scale and complexity to understand how exometabolites impact microbial community structure, soil health, and plant health.

Fabricated ecosystems enable discovery of gene and metabolite functions:  The EcoFAB 2.0 device is designed to grow and monitor small model plants under sterile conditions. Root and shoot growth, and plant exudate production can be tracked over time. Devices are automation compatible and can be modified for gas analyses including isotopic labeling. Since these devices are contained and controlled we can use plant and microbial mutants to test hypotheses for how metabolites mediate key processes in the rhizosphere. Our newest EcoFAB device is the EcoFAB 3.0 for growing larger bioenergy crops like Sorghum for at least a month in a sterile, controlled environment. Greenhouse experiments can be supplemented/expanded with the EcoFAB 3.0 to conveniently include: root growth visualization, isotopically labeling of plant biomass and exudates, perform metabolomics analyses under controlled conditions, safely assess mutant plants and microbes. Information from EcoFAB 3.0 devices can help feed AI/ML models including digital twins by Read K. Gupta et al. for more information.

Integrating AI, exometabolomics, and fabricated ecosystems using digital twins: Fabricated ecosystems enable control of key environmental and biological variables and measurement of plant, microbe, and dissolved organic responses. We are collaborating with many labs to integrate these capabilities in digital twin simulations refined using automated experimentation to both rapidly develop a molecular-level understanding of soil process and plant-microbe interactions

Using lab automation to accelerate research – Autonomous EcoBOT experiments: We have constructed the EcoBOT which automates the growth, sampling, and imaging of fabricated ecosystems. The EcoBOT holds 100s of EcoFAB 2.0 devices and uses a robotic arm to image, add and sample microbes, nutrients and other chemicals throughout growth. This allows us to conduct time-series studies to investigate the reciprocity between exometabolites, plant health, and microbiomes. As part of a collaboration with CAMERA (https://camera.lbl.gov/) we are integrating gpCAM software and AI computer vision to enable autonomous EcoBOT experiments designed to understand the role of microbes and metabolites in promoting soil and plant health.

Featured Software:
doi.org/10.1038/s41598-024-63497-8
https://gpcam.lbl.gov/
doi.org/10.1101/2024.12.20.629718

Artificial Intelligence and Informatics

Metabolite identification: BLINK (Blur-and-Link) is a Python package for efficiently generating cosine-based similarity scores and matching ion counts for large numbers of fragmentation mass spectra (Harwood et al., 2023). By bypassing fragment alignment and simultaneously scoring all pairs of spectra using sparse matrix operations, BLINK is over 3000 times faster than MatchMS, a widely used loop-based alignment and scoring implementation. Using a similarity cutoff of 0.7, BLINK and MatchMS had practically equivalent identification agreement, and greater than 99% of their scores and matching ion counts were identical. This performance improvement can enable calculations to be performed that would typically be limited by time and available computational resources.

SIMILE enables alignment of tandem mass spectra with statistical significance: Connecting small molecules through their aligned fragmentation spectra is a key aspect of tandem mass spectrometry-based untargeted metabolomics. Existing alignment algorithms often lack statistical significance and struggle with compounds that have multiple delocalized structural variations, leading to misalignment of fragment ions. Our approach, Significant Interrelation of MS/MS Ions via Laplacian Embedding (SIMILE), aligns fragmentation spectra with statistical significance while accommodating multiple chemical differences (Treen et al., 2022). SIMILE uncovers structural connections in molecular networks through spectral alignment that cosine-based scoring methods miss. Additionally, it enables ranking of spectral alignments using p-values, improving the exploration of structural relationships between compounds and enhancing the chemical connectivity in molecular networking.

MAGI: A Method for Metabolite Annotation and Gene Integration: Mass spectrometry features are connected to metabolites via methods such as accurate mass searching or fragmentation pattern matching. These metabolites are expanded to include similar metabolites by using the Chemical Network. These metabolites are then connected to reactions, which are reciprocally linked to input gene sequences via homology. The metabolite, reaction, and homology scores generated throughout the MAGI process are integrated to form MAGI scores (Erbilgin et al., 2019).

Bridging Knowledge Gaps: Computational Challenges and Opportunities in Microbiome Engineering with AI: From the current AI revolution, we anticipate having more accurate and detailed models of microbial growth, metabolism, and regulation; functional genomics will advance so that the majority of genes will no longer have no or unspecific annotations; and sensible frameworks will explain how microbial community sequence data explains the role and adaptability of members of these complex systems. Unfortunately, due to the complexity of most microbiomes, these advances may only improve specific outcomes for a limited range of applications. In areas where knowledge about mechanisms is sparse, even speculative assertions about causal relationships are challenging, and this limitation will likely be most prominent in terrestrial microbiomes. Consequently, it will remain a major challenge over the next decade to reliably engineer complex microbial systems to tackle the real-world problems mentioned above.

The Environmental Molecular Network (ENVnet): Dissolved organic matter (DOM) plays a crucial role in Earth’s soils but many chemical details remain poorly understood. In this study, we introduce ENVnet, a large-scale mass spectrometry-based molecular network of metabolites, aimed at deciphering the complexity of DOM across various environments. ENVnet reveals globally prevalent, structurally conserved molecules, offering fresh perspectives on DOM, especially long-lasting oxygenated alicyclic molecules. This tool allows researchers to test hypotheses regarding specific molecular structures in DOM related to building healthy soils and measuring soil health.

https://github.com/biorack/envnet