Transcription factors specifically control change

In this Outlook, Rothenberg discusses a new study in the previous issue of Genes & Development from Lo et al. that uses an elegant experimental system to directly compare the roles of transcription factor–binding site interaction in gene regulation maintenance with roles of the same factor–site interactions through developmental change.

Abstract

Transcription factors are defined by their sequence-specific binding to DNA and by their selective impacts on gene expression, depending on specific binding sites. The factor binding motifs in the DNA should thus represent a blueprint of regulatory logic, suggesting that transcription factor binding patterns on the genome (e.g., measured by ChIP-seq) should indicate which target genes the factors are directly controlling. However, although genetic data confirm high impacts of transcription factor perturbation in embryology, transcription factors bind to far more sites than the number of genes they dynamically regulate, when measured by direct perturbation in a given cell type. Also, deletion of carefully chosen transcription factor binding sites often gives disappointingly weak results. In a new study in the previous issue of Genes & Development, Lo and colleagues (pp. 1079–1095) reconcile these contradictions by using an elegant experimental system to directly compare the roles of transcription factor–binding site interaction in gene regulation maintenance with roles of the same factor–site interactions in gene regulation through developmental change. They examine Oct4:Sox2 shared target genes under maintained versus reinduced pluripotency conditions within the same cell clone. The results show that the same factor–site interaction impacts can appear modest in assays in developmental steady-state but are far more important as regulatory catalysts of developmental change.

Sequence-specific binding to regulatory elements in the DNA underlies the crucial roles that transcription factors exert in embryonic patterning and cell type specification, when cells must dynamically acquire new identities in a highly regulated, specific, and robust way (Spitz and Furlong 2012; Peter and Davidson 2015). By determining which transcription factor target sites were available, the genomic sequences at enhancer sites could thus be interpreted as a literal algorithm for developmental gene regulation (Peter and Davidson 2015). However, since the advent of ChIP-seq, it has been clear that transcription factor binding need not predict regulatory activity: A transcription factor can occupy 10 times to 100 times more binding sites across the genome than the number of genes that depend on that factor or respond to it in that cellular context (Biggin 2011; Revilla-i-Domingo et al. 2012). This has raised questions about the potency of sequence-specific transcription factors as causal drivers of gene regulation and about the functional importance of their individual binding sites. What is missing?

An important new study by Lo et al. (2022) indicates that the key is to ask the right question about transcription factor function. Their data show elegantly that the role of transcription factor binding to specific sites is much more crucial to change gene expression than it is to keep gene expression at a given stable level. Change is more sensitive than maintenance both for target genes that need to be activated and for genes that need to be repressed. We may thus be misunderstanding the true role of transcription factors when we assume that gene expression maintenance is a reflection of the same mechanisms as initiating gene activation or repression.

Lo et al. (2022) have used an experimental scheme of great clarity to compare maintenance and change-promoting roles of transcription factor binding. There are three elements to their strategy. As one key element, they have taken advantage of doxycycline-controllable induced pluripotent cells, developed by the group of Kathrin Plath (Sridharan et al. 2013). These provide a unique developmental state that is well defined but can be experimentally erased and then re-established at will. Individual clones of these cells can be fully characterized in their pluripotent cell state, then differentiated to neural precursor cells with complete loss of pluripotency, and then efficiently redifferentiated back to pluripotency over a well-defined time course. This strategy has enabled pluripotency-associated gene regulation to be compared within the same clone of cells both in the initial state, when it simply needs to be maintained, and in the “secondary” pluripotent state, when pluripotency is actively re-established. As a second strategic element, the investigators separated direct roles of the Oct4 and Sox2 transcription factors in controlling specific targets from any indirect roles that those transcription factors might play through their broader developmental impacts on the cells. After careful choice of targets, they performed their experimental mutagenesis on the specific binding sites at candidate regulatory elements rather than on the transcription factors themselves. Using this system, Lo et al. (2022) have shown that particular, highly specific sites that bind Oct4 and Sox2 in pluripotent cells may provide only minor quantitative effects on local gene expression levels in the context of stable pluripotency, but that the same sites can exert nearly all-or-none power over expression of the same local genes when the cell's pluripotency needs to be re-established.

A third element of strategy, much rarer in current functional genomics, was the investigators’ initial stratification of the genes to be studied. Rather than look at all expressed genes with binding sites for Oct4 and Sox2 together, Lo et al. (2022) first classified the expressed genes in the pluripotent cells into different bins, depending on how selectively they were expressed in a pluripotent context. They separated genes into a fully ES-specific, a “dynamic” (expressed in ES cells and, at most, a few other cell types), and a “broadly expressed” (expressed in ES cells and also in multiple other cell types) categories. They then compared the impacts of mutagenizing select, promoter-linked Oct4:Sox2 cobinding sites with high-specificity motifs, which were linked to representatives of each of these three gene categories. In fact, the strong impacts of site deletion during re-establishment of pluripotency were only seen for the ES-specific and dynamic genes—a striking justification of the strategy.

Mechanistically, this stratification could have been important for two reasons. A priori, redundancy among multiple enhancers for the same gene could mask the quantitative impact of mutating any individual enhancer (Kieffer-Kwon et al. 2013). Multienhancer organization is often associated with genes that are expressed in diverse biological contexts, but is less common with genes whose expression is confined to a single state (González et al. 2015). To be ES-specific or dynamic, genes may have had fewer diversely regulated enhancers contributing to their expression, thus increasing their dependence on particular enhancers served by Oct4 and Sox2. Another reason could have been that within the pluripotent cells themselves, the Oct4:Sox2 joint binding sites were actually much more common around ES-specific and dynamic genes than around broadly expressed genes. They were also much more associated with H3K27ac and with superenhancer type organization at sites where they were linked with ES-specific or with dynamic genes than in the sites where they were linked with broadly expressed genes, even though the broadly expressed genes might be expressed just as strongly. In this way, Oct4 and Sox2 were particularly good factors to follow compared with other transcription factors that typically show much higher levels of binding to random, well-expressed, but developmentally uncorrelated, genes. Thus, by leveraging the high selectivity of this particular factor combination and by distinguishing the developmental classes of genes where these factors bound, the investigators began with a good chance to enrich for sites that would strongly regulate genes with Oct4- and Sox2-dependent expression.

Indeed, these criteria enhanced the opportunity to see a strong effect of the key Oct4:Sox2 binding sites at ES-specific and dynamic genes whenever it occurred—if it occurred. This underscored the difference between the weak (approximately two times) effect of mutating these binding sites on ES-specific and dynamic gene expression, seen during pluripotent state maintenance (comparable with results in other systems), and the powerful, all-or-none impact of the same sites on expression of the same genes within the same cell clones, when the cells re-established pluripotency during secondary reprogramming.

These elegant experiments are artisan-scale, not global. However, they provide a way to reconcile decades of evidence for transcription factor power in developmental genetics and in acute immune or inflammatory gene activation systems, on the one hand, with more recently accumulated evidence for “inertial” mechanisms on the other hand: mechanisms like reader–writer chromatin modification propagation, superenhancer organization, and liquid–liquid phase separation condensates, all of which would be expected to resist gene expression change generally (Allis and Jenuwein 2016; Sabari et al. 2018). Lo et al. (2022) could argue that these latter gene regulation mechanisms are doing a very different job for multicellular organisms than the job of sequence-specific transcription factors binding to specific enhancers. Their results show that transcription factors acting through specific sites can be far more potent as catalysts of change than as caretakers of differential gene regulation states.