Causal dynamical modelling predicts novel regulatory genes of FOXP3 in human regulatory T cells

Rucha Sawlekar, Stefano Magni, Christophe Capelle, Alexandre Baron, Ni Zeng, Laurent Mombaerts, Zuogong Yue, Ye Yuan, Feng He, Jorge M.s. GonçalvesbioRxiv 2020 · DOI

TL;DR

This paper introduces a computational approach to identify novel upstream regulators of FOXP3, a key transcription factor in regulatory T cells (Tregs), by modeling causal relationships from time-series gene expression data. Experimental validation confirmed that several predicted regulators, notably NRBF2, significantly impact FOXP3 expression at both mRNA and protein levels, offering new insights into immune regulation and disease mechanisms.

What Problem Does This Paper Address?

Regulatory T cells (Tregs) are crucial for maintaining immune tolerance and preventing autoimmune diseases. The transcription factor FOXP3 is central to Treg function, but the upstream genetic machinery that controls FOXP3 itself remains largely unknown. Identifying these regulators is vital for developing new immunotherapies. Traditional methods for discovering these regulators are often laborious and inefficient when screening the entire genome. This paper aims to bridge this gap by using computational modeling to predict potential FOXP3 regulators from large-scale gene expression data, followed by experimental validation.

Methods

The core of the methodology is a "causal dynamical modeling" approach using time-series gene expression data. Specifically, they employed a "one-to-one" and "all-to-one" system identification strategy based on linear time-invariant (LTI) models.

  • Time-Series Data Normalization and Filtering: Raw microarray data from human Tregs was pre-processed using the germa algorithm for normalization. Subsequent filtering removed transcripts with low expression or inconsistent data across measurements.
  • System Identification (One-to-One): For each potential regulator transcript, a first-order LTI model was fitted to predict FOXP3 expression. This model captures the causal relationship between the regulator and FOXP3, considering both direct effects and potential time delays. The fitness of each model was assessed to rank the potential regulators.
  • Incorporating Time Delays: To account for biological processes that introduce delays between a regulator's action and its target's response, the models were extended to include time delays (0-100 minutes) in the input signal.
  • All-to-One Approach: This stage combined the results from multiple potential regulators (all transcripts) predicting a single target (FOXP3). The highest fitness score across different time delays was used to rank the input transcripts.
  • Selection for Validation: After ranking, a subset of top-ranked genes was selected for experimental validation based on fitness scores, dynamic profiles, and biological relevance criteria.

Datasets

The study utilized time-series genome-scale mRNA expression data from isolated human regulatory T cells (Tregs) obtained from two different donors. These Tregs were stimulated with anti-CD3/-CD28/IL2, and measurements were taken at 20-minute intervals over a six-hour period. After pre-processing, the dataset comprised 13,601 common transcripts.

Key Insights and Ideas

  • Computational Prediction of Regulatory Networks: The paper demonstrates the power of dynamical modeling on time-series data to infer causal relationships, moving beyond simple correlation to identify potential upstream regulators of a key transcription factor like FOXP3.
  • Importance of Time Delays: Explicitly modeling time delays in gene regulation is crucial for accurately capturing the dynamics of biological systems and identifying relevant interactions.
  • NRBF2 as a Novel Regulator: The identification and experimental validation of NRBF2 as a regulator of FOXP3 is a significant finding, opening new avenues for understanding immune homeostasis and disease.
  • mRNA vs. Protein Level Effects: The study highlights that while mRNA expression is a good starting point, validating effects at the protein level (as done for NRBF2) is essential for a complete understanding of regulatory mechanisms.

Key Results

  • The computational pipeline successfully reduced the initial ~13,601 transcripts to a ranked list of potential FOXP3 regulators.
  • Out of five selected candidate genes (NRBF2, NCOA7, RNF12, PDE4D, and MAP1LC3B) for experimental validation, three (MAP1LC3B, NCOA7, and NRBF2) showed a significant effect on FOXP3 mRNA expression upon knockdown.
  • Further experiments confirmed that NRBF2 knockdown not only reduced FOXP3 mRNA but also affected FOXP3 protein expression, suggesting a role in post-transcriptional regulation or alternative splicing.

Limitations and Future Work

  • Linear Model Simplification: The authors acknowledge that using linear models might miss complex, non-linear interactions between genes.
  • Single Transcript-at-a-Time Approach: The method initially considered each transcript individually, potentially missing synergistic effects where multiple regulators act in concert.
  • Limited Experimental Resources: The selection of only five genes for validation was constrained by available resources, suggesting that many other predicted regulators might exist.

Future work suggested by the authors includes:

  • In vivo validation of the identified regulators (NRBF2, NCOA7, RNF12) in animal models.
  • Experimental investigation of additional top-ranked genes from the predicted list.
  • Applying the same computational approach to identify regulators of other key immune genes like CTLA4.
  • Developing and applying more complex models that can capture non-linear interactions and mechanistic details, especially with the availability of new data types like single-cell RNA-seq.

Summary updated

Abstract

Abstract Regulatory T cells (Tregs), characterized as a CD4+CD25+FOXP3+ subset of T cells, are vital to the induction of immune tolerance and the maintenance of immune homeostasis. While target genes of Treg master regulator FOXP3 have been identified, the upstream regulatory machinery of FOXP3 still remains largely unknown. Here we dynamically model causal relationships among genes from available time-series genome-scale datasets, to predict direct or indirect regulatory genes of FOXP3 in human primary Tregs. From the whole genome, we selected five top ranked candidates for further experimental validation. Following knockdown, three out of the five candidates indeed showed significant effects on the mRNA expression of FOXP3. Further experiments showed that one out of these three predicted candidates, namely nuclear receptor binding factor 2 (NRBF2), also affected FOXP3 protein expression. These results open new doors to identify potential new mechanisms of immune related diseases.
BibTeX
@misc{sawlekar2020causal,
  author = {Rucha Sawlekar and Stefano Magni and Christophe Capelle and Alexandre Baron and Ni Zeng and Laurent Mombaerts and Zuogong Yue and Ye Yuan and Feng He and Jorge M.S. Gonçalves},
  title = {{Causal dynamical modelling predicts novel regulatory genes of FOXP3 in human regulatory T cells}},
  year = {2020},
  doi = {10.1101/2020.02.13.943688},
  howpublished = {bioRxiv},
}
Figure