Transcription factors are the master switches of the genome. In humans, roughly 1,600 of these proteins read the DNA sequence and decide, in concert with one another, which of our roughly 20,000 genes are active in any given cell. They do this by recognizing specific short stretches of DNA — typically 6 to 12 base pairs — known as binding motifs. Knowing exactly which sequence each factor prefers is foundational to nearly everything in genomics: predicting how genetic variants contribute to disease, engineering synthetic gene circuits, interpreting genome-wide association studies, and understanding how mutations in regulatory regions drive cancer. Yet despite the importance of these measurements, the standard methods used to derive them have long been recognized as imperfect.

The dominant approaches — including protein binding microarrays, conventional SELEX (Systematic Evolution of Ligands by Exponential Enrichment), and various high-throughput SELEX variants — share a common limitation: they typically measure binding under a single, fixed set of conditions, and they often rely on initial libraries whose sequence diversity constrains what can be discovered. A transcription factor that binds weakly, binds cooperatively, or requires specific flanking context can easily be mischaracterized. GHT-SELEX, the method introduced and deployed in the new study, was designed to close these gaps. The “GHT” in its name reflects an expanded high-throughput design that dramatically increases both the depth of sequencing and the diversity of the DNA library interrogated in each round of selection, allowing the technique to capture binding events that earlier methods would have missed entirely.

In a GHT-SELEX experiment, a vast library of random DNA sequences is incubated with a purified transcription factor. The sequences that the protein binds are separated from those it ignores — typically using techniques that pull down the protein-DNA complexes — and the bound sequences are then amplified and subjected to another round of selection. After several iterative cycles, the pool becomes enriched for high-affinity binding sequences, and deep sequencing reveals what the protein “chose.” The statistical analysis of these enriched sequences, combined with careful modeling of binding energies, allows researchers to reconstruct a precise portrait of the protein’s sequence preferences, including subtle dependencies between positions that simpler models cannot capture. The key innovation in this study lies in scaling this approach up and in applying it systematically to a large panel of human transcription factors under conditions designed to reveal their full behavioral repertoire.

The results were striking. A substantial number of the transcription factors examined displayed sequence specificity that is “unexpectedly high” — meaning their discrimination between favored and disfavored DNA sequences is far sharper than earlier assays had indicated. Where previous studies might have characterized a factor as recognizing a loose, degenerate motif, GHT-SELEX revealed that the protein in fact distinguishes finely between closely related sequences, tolerating only a narrow band of variation. This matters enormously for interpretation of genomic data. If a factor is actually highly specific but has been modeled as promiscuous, then computational predictions of where it binds across the genome — and which genetic variants might disrupt those bindings — will be systematically wrong.

Just as consequential is the second headline finding: complex DNA binding. Many of the factors did not behave as simple, independent-position recognizers at all. Instead, their binding depended on interactions between positions in the motif, on the spacing and orientation of multiple recognition elements, and in some cases on the ability to engage more than one DNA site at a time. Some factors showed evidence of dimeric binding on concatenated sites; others exhibited context-dependent preferences in which the sequence flanking the core motif altered what the core itself could be. These are exactly the kinds of behaviors that conventional motif models — which assume each position in a binding site contributes independently to binding affinity — cannot represent. The study’s data indicate that such “independent position” assumptions, baked into the position weight matrix models used ubiquitously in bioinformatics, fail for a meaningful fraction of human transcription factors.

The implications ripple outward across several fields. In regulatory genomics, motif scanning underlies algorithms such as those used to annotate transcription factor binding sites in the human genome and to interpret ENCODE-style functional element catalogs. If the underlying specificity models are too coarse, then hundreds of thousands of predicted binding sites may be false positives, while genuinely important sites — those that depend on complex, cooperative or context-sensitive binding — may be absent from the catalogs entirely. In medical genetics, fine-mapping studies that try to pinpoint which regulatory variant explains a disease association rely on motif disruption scores; sharper, more accurate specificity models should translate directly into better variant interpretation. In synthetic biology, engineers who design genetic circuits using natural transcription factors will now have better ground truth about what sequences their parts actually respond to.

The methodological advance is itself noteworthy. By combining an enlarged library design with high-depth sequencing and quantitative modeling, GHT-SELEX achieves a dynamic range that allows both very strong and comparatively weak binding interactions to be measured within the same experiment. This is important because biological reality rarely fits a single affinity: transcription factors in living cells encounter a spectrum of sites, and their functional occupancy depends on affinity, competition with other factors, chromatin context and concentration. Having an in vitro assay that can resolve this spectrum — rather than collapsing it into a single consensus motif — brings the measurement considerably closer to the biology.

The study also underscores how much of the human transcription factor repertoire remains incompletely characterized. Even for well-studied families — the homeodomains, bZIPs, bHLHs, nuclear receptors and zinc finger proteins that appear throughout the gene regulation literature — the new measurements revealed surprises, including preferences and binding modes not captured in existing databases such as JASPAR or HOCOMOCO. For factors previously annotated only by similarity to relatives, the new data provide direct experimental characterization, some of it quite different from what homology-based transfer would have predicted. The authors’ systematic approach, applying the same protocol across a broad panel of proteins, also makes the resulting dataset unusually consistent and comparable — a valuable resource for anyone building predictive models of gene regulation.

For the broader field, the paper is likely to prompt a re-evaluation of how transcription factor specificity is measured and modeled. Position weight matrices will not disappear overnight — they remain useful first approximations — but the study strengthens the case for higher-order models, such as dinucleotide models and deep-learning-based approaches, that can capture inter-positional dependencies. It also makes a strong argument for experimental rigor: the intrinsic preferences of a transcription factor, measured cleanly in vitro with sufficient dynamic range, can differ enough from legacy measurements to change biological conclusions. As researchers begin incorporating the new specificity data into genome-wide analyses, one thing seems certain: the grammar of gene regulation, long treated as a simple code of short motifs, is proving to be considerably richer — and considerably more precise — than the textbooks suggested.

Modern genetic studies can identify hundreds or even thousands of DNA variants associated with conditions such as asthma. Yet an associated variant is not necessarily located within the gene responsible for disease, nor does it always reveal how the disease develops. Many variants influence gene activity from a distance, while several nearby genes may be affected by the same regulatory region. This makes it difficult to determine which genes are genuine disease drivers and which are simply passengers in a complex genetic signal.

DANDELION was developed to address this problem through a mediation-inspired analytical strategy. Rather than treating a genetic variant and a disease outcome as directly connected, the framework examines whether the variant may influence disease through changes in gene regulation. It incorporates trans-regulatory information, meaning regulatory effects that can operate across genomic regions, and integrates these data with genetic associations observed in large-scale studies. By tracing potential regulatory routes from DNA variation to gene activity and then to disease, the method seeks to distinguish biologically meaningful drivers from genes that are only indirectly associated with risk.

“Current genetic approaches and large-scale studies often identify hundreds of DNA changes linked to disease risk, but it can be difficult to determine which genes are actually causing the disease rather than simply being associated with it,” said Zhonghua Liu, ScD, assistant professor of biostatistics at Columbia University Mailman School of Public Health. According to the researchers, DANDELION offers a way to rank candidate genes according to their likely role in disease biology, potentially making genetic discoveries more useful for therapeutic development.

The team applied the framework to asthma, a chronic respiratory disease characterized by airway inflammation, variable airflow obstruction and heightened sensitivity to environmental triggers. After identifying candidate disease-driving genes, the researchers compared their results with single-cell gene-expression data from the Human Lung Cell Atlas. This allowed them to determine which individual lung cell types showed the strongest activity of the prioritized genes, providing a more precise view of where the implicated biological processes may operate.

The analysis highlighted protein palmitoylation, a chemical modification that attaches fatty acids to proteins, as a potentially important pathway in asthma. Palmitoylation can alter a protein’s stability, movement, interactions with other molecules and positioning within a cell. Because these functions influence signaling and immune activity, disruptions in palmitoylation could affect the behavior of cells involved in airway inflammation. The researchers’ findings suggest that this pathway may contribute to asthma in ways not captured by conventional gene-prioritization methods.

Laboratory experiments provided additional support for the computational results. The investigators found that enzymes involved in protein palmitoylation influenced inflammation-related processes associated with asthma. These experiments do not establish that the enzymes can be safely targeted in patients, but they strengthen the case that palmitoylation is biologically connected to the disease rather than being a statistical signal alone. The pathway could therefore serve as a starting point for future studies seeking treatments that modify disease mechanisms at the cellular level.

The study was co-led by Zhonghua Liu, Marcelo A. Nóbrega, MD, PhD, professor in the Department of Human Genetics at the University of Chicago, and Xuanyao Liu, PhD, assistant professor in the Departments of Medicine and Human Genetics at the University of Chicago. The investigators describe DANDELION as a general framework rather than an asthma-specific tool. Because it can integrate genetic data with regulatory information from relevant tissues, it may be applicable to other complex diseases in which multiple variants, cell types and biological pathways contribute to risk.

The researchers say the approach could help narrow the gap between genetic discovery and drug development. Genetic evidence is increasingly used to select potential therapeutic targets, but the large number of associations produced by genome-wide studies can make that process difficult. By highlighting genes that appear to mediate the effects of disease-associated variation, DANDELION may provide a more focused list of candidates for functional experiments and clinical investigation. The software and analysis code have been made publicly available, allowing other researchers to test the framework in additional diseases and datasets. The work was supported by the National Institutes of Health and the American Heart Association, including NIH grant R01AG086379 to Zhonghua Liu. The authors reported no competing interests.

Subject of Research: Genetic drivers and biological mechanisms of asthma

Article Title: Trans-regulatory gene mapping prioritizes disease drivers in asthma

Web References: https://www.github.com/mxxptian/DANDELION; https://www.mailman.columbia.edu/

References: Cell; DOI: 10.5281/zenodo.19911608

Keywords: DANDELION, asthma, disease-driving genes, genetic analysis, trans-regulatory mapping, protein palmitoylation, lung inflammation, Human Lung Cell Atlas, therapeutic targets, computational biology