TL;DR
This paper introduces MetaGP, a statistical method to identify activated transcription factors (TFs) from gene expression data, providing a "synoptic map" of TF activity. They apply this to study the effects of Kampo medicine on demyelination in mice, offering a systems-level view of complex herbal drug mechanisms.
What Problem Does This Paper Address?
Understanding how biological systems respond to stimuli, especially complex ones like herbal medicines, is challenging. Transcription factors (TFs) are key regulators of gene expression, but their direct activity (TFAs) is hard to measure. Gene expression data, while abundant, doesn't always directly reflect TF activity, as TF gene expression levels can be low and indirect. Existing methods often struggle to provide a holistic view, especially when dealing with drugs composed of multiple active compounds, like Kampo medicines. This paper aims to bridge this gap by developing a method to infer TFAs from gene expression data and apply it to a complex biological scenario.
Methods
The core of the paper is the Meta Gene Profiler (MetaGP) method. It's a statistical test designed to evaluate the significance of a set of genes, rather than individual genes. This is crucial because TFs regulate multiple genes, and their collective effect is more informative than individual gene expression changes.
Here's how it works:
- Gene Set Definition: Genes are grouped based on annotations. In this study, these groups represent genes regulated by a specific TF (identified using TRANSFAC database for promoter binding sites) or genes belonging to a Gene Ontology (GO) term.
- Individual Gene Testing: For each gene within a set, a statistical test (Welch's t-test) is performed to compare gene expression between case and control groups across different time points. This yields a p-value for each gene.
- Meta-Analysis (Normal Inversion): MetaGP then integrates these individual gene p-values using a technique called "normal inversion." This method transforms the p-values into z-scores, which are then combined to produce a single, integrated p-value for the entire gene set (e.g., all genes regulated by a specific TF).
- Absolute Evaluation: A key characteristic of MetaGP is its "statistical absolute evaluation." Unlike rank-based methods, MetaGP evaluates statistical evidence independently of other gene sets. This allows for meaningful comparisons of TFAs across different experimental conditions (e.g., different treatments or time points) using the same standard.
Datasets
The primary dataset used is microarray gene expression data from a case-control experiment involving mice treated with Kampo medicine.
- Experimental Setup: Mice were subjected to different conditions:
- E1 (Control): Normal diet.
- E2 (NYT): Dosed with 1% Ninjin'yoeito (NYT), a Kampo medicine.
- E3 (Cuprizone): Dosed with 0.2% cuprizone to induce demyelination for 5 weeks, followed by a normal diet.
- E4 (Cuprizone + NYT): Dosed with cuprizone for 5 weeks, then NYT was continued.
- Sampling: Mice were sacrificed at the 3rd and 7th weeks.
- Annotation Data:
- TF annotations: Derived from the TRANSFAC database, identifying genes with TF binding sites in their promoter regions (306 unique TFs).
- GO terms: Predefined gene sets from the Gene Ontology (BP, CC, MF) were also used (over 20,000 terms).
Key Insights and Ideas
- Holistic View of Herbal Medicine: Kampo medicines are complex mixtures. MetaGP's ability to provide a "synoptic map" of TF activities is crucial for understanding the system-level effects of such drugs, moving beyond single-compound analysis.
- Inferring Activity, Not Just Expression: The paper highlights that inferring TF activity from gene expression is more informative than just looking at TF gene expression levels, which can be misleading.
- Absolute Evaluation for Comparative Studies: The "absolute evaluation" nature of MetaGP is a significant advantage, enabling direct comparison of TF activities across different experimental conditions, which is essential for understanding drug effects over time and dosage.
- Identifying Interaction Effects: The analysis of specific TFs (like RFX) suggests that MetaGP can reveal synergistic or antagonistic interactions between different components of a Kampo medicine or between the medicine and disease-inducing agents.
Key Results
- Synoptic Map of TF Activities: The study successfully generated a heatmap (Fig. 1) visualizing the integrated p-values (indicating TF activation) for 306 TFs across different experimental conditions and time points.
- Differential TF Activation: The analysis revealed that TFs were differentially activated in cuprizone-dosed and cuprizone-NYT-dosed mice, suggesting that cuprizone plays a significant role in inducing these TFAs. NYT-dosed cases alone did not show significant TF activation.
- Specific TF Insights:
- RFX: Activated in the cuprizone + NYT group (E4) but not in cuprizone (E3) or NYT (E2) alone, suggesting an interaction effect.
- NF-κB and IRF-1: Showed similar patterns in E3 and E4, indicating cuprizone's primary influence, with NYT not significantly altering this effect.
- STAT1: Exhibited a time effect, with activation in E3 at the 7th week but not E4, suggesting NYT might repress cuprizone's effect and stabilize the system.
- USF and CREB: Also showed time effects, with NYT's interaction effect for USF and a repressing effect for CREB at the later time point.
- GO Term Validation: The method was also applied to GO terms, showing significant activation for terms related to "myelination" and "structural constituent of myelin sheath," validating the approach for biological processes relevant to the study.
Limitations and Future Work
- TF Annotation Accuracy: The authors acknowledge that the annotation of genes to TFs based solely on promoter binding sites (from TRANSFAC) is not perfect. It's possible that not all identified genes are truly regulated by the TF. They suggest that incorporating data from ChIP-chip experiments could improve accuracy.
- Limited GO Terms Shown: Due to space limitations, only two GO terms related to myelination were presented, despite analyzing over 20,000. A more comprehensive analysis of GO terms could yield further insights.
- Future Directions: The authors suggest that investigating specific TFs and their corresponding gene expressions systematically can lead to newer insights into drug mechanisms and disease processes. They also emphasize that MetaGP, combined with other annotations like GO terms, provides a powerful tool for gaining a holistic understanding of biological systems, which is particularly valuable for complex interventions like Kampo medicine.
Summary updated