Abstract
Transcription factor (TF) activity serves as a critical readout of cellular signaling pathways, yet existing technologies often fail to assess activity changes across multiple TFs simultaneously or determine whether a TF acts as a global activator or repressor in a specific context. We present diffTF, a genome-wide method that integrates chromatin accessibility data (ATAC-Seq or DNase-Seq) with RNA-Seq and in-silico predicted binding sites. By aggregating signals across thousands of genomic locations, diffTF quantifies differential TF activity and classifies TFs by their mode of action. We validated this approach using chronic lymphocytic leukemia (CLL) patient samples and hematopoietic differentiation models, identifying both known and novel regulatory drivers.
Introduction
Transcription factors are the primary orchestrators of gene expression, translating cellular signals into genomic responses. Despite their importance, studying TFs at a systems level remains challenging. Traditional proteomics often lacks the sensitivity to detect low-abundance TFs, while ChIP-Seq—the gold standard for mapping TF binding—is resource-intensive and typically restricted to one factor at a time. Furthermore, the functional role of a TF is often context-dependent; a single protein may act as an activator in one cell type and a repressor in another.
To address these limitations, we developed diffTF, a computational framework designed to provide a global view of TF activity and classify their regulatory modes using widely available multiomics data.
The diffTF Framework
The diffTF pipeline operates through the integration of chromatin accessibility and transcriptomic data across two primary modes:
1. Differential Activity (Basic Mode)
In its basic mode, diffTF estimates the differential activity of hundreds of TFs by mapping chromatin accessibility reads onto thousands of predicted TF binding sites (TFBS). By calculating the fold-change in accessibility between conditions and comparing it against a background of GC-matched genomic regions, the tool identifies TFs with significant shifts in global occupancy. A key strength of this approach is its reliance on aggregate signals; while accessibility changes at individual sites may be subtle or statistically insignificant, the combined signal across all predicted sites for a specific TF provides a robust measure of activity.
2. Classification of Regulatory Mode (Multiomics Mode)
The classification mode integrates matched RNA-Seq data to determine a TF’s mode of action. By correlating the expression levels of a TF with the accessibility of its target sites across individuals or replicates, diffTF distinguishes between:
- Activators: TFs that show a positive correlation between their own expression and the accessibility of their binding sites.
- Repressors: TFs that show a negative correlation, where increased TF abundance leads to decreased accessibility at target sites.
Additionally, diffTF employs genomic footprinting to analyze the precise DNA "shadows" left by TFs. This allows for the identification of specific molecular signatures, such as the observation that repressors may increase accessibility at the immediate binding site while causing compaction in the surrounding 10–100 base pairs.
Case Studies and Validation
Chronic Lymphocytic Leukemia (CLL)
We applied diffTF to a large dataset of over 50 patients, comparing mutated (M-CLL) and unmutated (U-CLL) samples. The analysis identified 67 differentially active TFs. While 40% of these were previously associated with CLL, 60% represented potentially novel drivers, including members of the circadian clock family such as BMAL1 and CLOCK. We successfully classified approximately 44% of the expressed TFs as either activators or repressors, providing a more nuanced map of the CLL regulatory landscape.
Hematopoietic Differentiation
To test the tool's performance on smaller datasets, we analyzed murine multipotent progenitors (MPP) and granulocyte-macrophage progenitors (GMP). Despite having only four replicates per condition, diffTF correctly identified the CEBP family as the dominant drivers of myeloid differentiation. The majority of differentially active TFs in this context were classified as transcriptional activators.
Discussion and Insights
Our findings highlight the importance of data-driven classification over static database annotations. For instance, PAX5—often broadly categorized in literature—was identified as a clear repressor in the context of CLL. This contextual plasticity underscores the need for tools like diffTF that evaluate TF function within the specific biological environment under study.
Furthermore, diffTF demonstrated remarkable robustness. It remains accurate even when using in-silico predicted binding sites, which are known to contain false positives, and performs reliably at low sequencing depths (1–5 million reads).
Limitations
While powerful, diffTF has limitations. It currently assigns a global "average" mode of action, which may obscure the behavior of bifunctional TFs that act as activators at some loci and repressors at others. Additionally, the reliance on RNA levels as a proxy for TF abundance may overlook factors regulated primarily through post-translational modifications. Finally, the genome-wide analysis of hundreds of TFs is computationally intensive, requiring significant memory and processing time.
Conclusion
diffTF provides a scalable, high-resolution method for quantifying TF activity and determining regulatory modes. By bridging the gap between chromatin accessibility and gene expression, it offers a powerful resource for identifying the key regulatory drivers of cellular identity and disease progression.
The diffTF software is open-source and available at: https://git.embl.de/grp-zaugg/diffTF
Summary updated