TL;DR
This paper introduces a novel computational framework to untangle the complex causal relationships between metabolism, signaling, and gene regulation in living cells. By integrating dynamic multi-omics data with prior biological knowledge, the researchers identified key metabolic signals that influence the TORC1 pathway in yeast, including a strong candidate for the long-hypothesized glutamine signal.
What Problem Does This Paper Address?
Understanding how cells respond to their environment is fundamental to biology. This response involves intricate networks of signaling molecules, metabolic pathways, and gene expression. However, these networks are often characterized by complex feedback loops, making it difficult to pinpoint which metabolic signals are truly driving cellular changes. Specifically, the endogenous inputs that trigger nutrient signaling pathways, like the crucial TORC1 pathway, remain largely unknown. While experimental techniques can now generate vast amounts of dynamic data across different molecular levels (metabolites, proteins, transcripts), integrating this data to infer causal relationships, especially in the face of feedback, is a significant challenge. This paper aims to bridge this gap by developing a method to infer these causal links from dynamic, multi-level data.
Methods
The core of this work is a novel probabilistic, model-based inference framework. This framework is designed to infer causal relationships by integrating dynamic multi-omics data (transcriptomics, metabolomics) with prior biological knowledge.
Key methodological components include:
- Co-designed Dynamic Experiments: The researchers orchestrated specific experimental conditions (nitrogen source shifts and rapamycin treatment) to generate dynamic data that could be systematically analyzed. This "perturbation matrix" approach is crucial for inferring causality.
- Network Motifs: The framework categorizes potential metabolic signals into four "network motifs" that describe different modes of interaction with signaling pathways and gene expression: "unrelated," "downstream" (metabolite changes are a consequence of signaling), "upstream" (metabolite changes are signals to the pathway), and "parallel" (metabolite and gene expression changes occur independently).
- Feature Extraction: For each metabolite-transcript pair, two key features are computed:
- Dynamic Dependence (DD): This quantifies how consistently a metabolite's change over time predicts the change in a transcript's expression, accounting for time delays.
- Representation of TOR Genes (RG): This measures how strongly a metabolite's associated genes are known targets of TORC1-regulated transcription factors.
- Bayesian Inference: These features are then fed into a Bayesian framework to calculate the probability of a metabolite belonging to each of the four network motifs. This probabilistic approach allows for robust assignments and quantification of uncertainty.
- Robustness Analysis: The authors rigorously tested the reliability of their assignments by varying assumptions about prior knowledge (e.g., the number of known TORC1-controlled transcription factors) and computational parameters.
Datasets
The study generated new, comprehensive dynamic datasets from Saccharomyces cerevisiae (budding yeast) under three distinct experimental conditions:
- Nitrogen Upshift: Shifting from proline to glutamine as the sole nitrogen source.
- Nitrogen Downshift: Shifting from glutamine to proline.
- Chemical Downshift: Inhibiting the TORC1 pathway directly with rapamycin while growing on glutamine.
These experiments involved:
- Transcriptomic data: Measuring the abundance of 5,716 transcripts.
- Proteomic data: Quantifying 20 TOR-related signaling proteins (though these were found to be largely constant).
- Metabolomic data: Measuring 42 intracellular metabolites using targeted LC-MS/MS and an additional 256 using untargeted FIA-QTOF-MS.
The data was collected at high temporal resolution (up to 15 samples per shift) across three independent biological replicates.
Key Insights and Ideas
- The "Upstream" Motif is Key for Signal Discovery: The framework's ability to identify metabolites acting "upstream" of a signaling pathway is crucial for discovering endogenous signals, a long-standing challenge.
- Glutamine as a TORC1 Signal is Quantitatively Supported: The study provides strong computational evidence for glutamine acting as an upstream signal to TORC1, validating previous hypotheses.
- AICAR as a Novel Candidate Signal: The metabolite AICAR, involved in histidine and purine metabolism, emerged as another promising "upstream" candidate, suggesting a potential link between nitrogen availability and TORC1 regulation via this molecule.
- TORC1 Influences Multiple Metabolic Processes: The identified "downstream" metabolites (e.g., trehalose-6-phosphate, IMP, adenosine) highlight TORC1's role in regulating carbohydrate storage, nucleotide salvage, and amino acid biosynthesis through post-translational modifications.
- Rapamycin Mimics Natural Downshifts: The similarity in transcriptomic responses between the rapamycin-induced and nutrient downshifts confirms that rapamycin is a valid tool for studying TORC1 inactivation in a controlled manner.
- Computational Framework is Generalizable: While applied to yeast and TORC1, the underlying methodology of integrating dynamic multi-omics data with network motifs and Bayesian inference is adaptable to other signaling pathways and organisms.
Key Results
- The computational framework successfully assigned metabolites to four distinct network motifs ("unrelated," "downstream," "upstream," "parallel").
- Six metabolites were identified as potential "upstream" signals to TORC1, including glutamine (highly supported) and AICAR (novel candidate).
- Seven metabolites were classified as "downstream" of TORC1, indicating their changes are a consequence of TORC1 activity, such as trehalose-6-phosphate (T6P) and purine-related metabolites (IMP, adenosine).
- Seven metabolites were identified as "parallel" signals, suggesting they influence transcription independently of TORC1.
- The robustness analysis demonstrated that the assignments were generally stable, even with variations in prior knowledge about TORC1-controlled transcription factors.
Limitations and Future Work
- "Hybrid" Motifs: The current framework might miss metabolites that play multiple roles (e.g., acting as both an upstream signal and a downstream product). Future work could explore more complex "hybrid" motif assignments.
- Prior Knowledge Dependency: While robust, the accuracy of assignments is still influenced by the completeness of prior knowledge, particularly regarding TORC1-controlled transcription factors.
- Protein Data Integration: The study noted constant protein levels, limiting direct integration of dynamic proteomic data. Future work could incorporate dynamic (phospho-)proteomics to unravel regulation across more network types and timescales.
- Mechanistic Details: While the framework infers causal relationships, it doesn't fully elucidate the precise biochemical mechanisms. Incorporating more detailed mechanistic information (e.g., metabolic network structures, enzyme kinetics) could further refine predictions.
- Applicability to Other Systems: While the framework is generalizable, the specific formulation of prior knowledge (network motifs, prototypic interactions) is application-specific. Adapting it to other signaling networks will require careful definition of these elements.
- Higher Eukaryotes: The authors suggest the approach could be applicable to higher cells if transcriptional or phosphorylation network topologies are known, opening avenues for studying human diseases.
Summary updated