The Non-continuum Nature of Eukaryotic Transcriptional Regulation

Abstract

Eukaryotic transcription factors are versatile mediators of specificity in gene regulation. This versatility is achieved through mutual specification by context-specific DNA binding on the one hand, and identity-specific protein-protein partnerships on the other. This interactivity, known as combinatorial control, enables a repertoire of complex transcriptional outputs that are qualitatively disjoint, or non-continuum, with respect to binding affinity. This feature contrasts starkly with prokaryotic gene regulators, whose activities in general vary quantitatively in step with binding affinity. Biophysical studies on prokaryotic model systems and more recent investigations on transcription factors highlight an important role for folded state dynamics and molecular hydration in protein/DNA recognition. Analysis of molecular models of combinatorial control and recent literature in low-affinity gene regulation suggest that transcription factors harbor unique conformational dynamics that are inaccessible or unused by prokaryotic DNA-binding proteins. Thus, understanding the intrinsic dynamics involved in DNA binding and co-regulator recruitment appears to be a key to understanding how transcription factors mediate non-continuum outcomes in eukaryotic gene expression, and how such capability might have evolved from ancient, structurally conserved counterparts.

1. Introduction

Complex multicellular organisms rely on a highly adaptive and finely tuned physiology to develop and survive. Many of the evolutionary innovations found in eukaryotes function at the transcriptional level and are physically mediated by DNA-binding transcription factor proteins. In single-celled prokaryotes such as bacteria, transcriptional repressors and activators directly interact with metabolites which modulate their binding affinity to target operators and regulate transcriptional output in an affinity-dependent manner. In eukaryotic organisms, gene expression must additionally be spatially coordinated with respect to cell type and body plan, as well as temporally coordinated with respect to homeostatic setpoints. In both prokaryotes and eukaryotes, the condensation of genomic DNA into chromatin represents a broad layer of regulation that operates over large clusters of genes at the same time (Szabo et al. 2019). The eukaryotic chromatin exhibits significant complexity in topology as well as DNA-modifying chemistry over their prokaryotic counterparts (Sikorska and Sexton 2020; Willenbrock and Ussery 2004; Li et al. 2011). However, transcriptional complexity in eukaryotes in the mesoscale is also accompanied by complexity at the molecular level. One needs but a glance at the array of transcription factors in the IFN-β enhanceosome (Panne et al. 2007; Panne 2008) to appreciate the complex machinery that regulates eukaryotic transcription (Fig. 1).

Fig. 1. The interferon-β enhanceosome exemplifies combinatorial control.

An atomic model of three transcription factors — interferon response factors NF-κB (green), IRF3/IRF7 (orange), and ATF-2/c-Jun (purple) — arrayed cooperatively to the IFN-β enhancer. Formation of this enhanceosome stimulates the transcription of the IFN-β gene. In this structure, virtually every base position is contacted by the bound factors

To meet the complex needs of multicellular physiology, eukaryotic proteins possess distinct molecular features that are absent in prokaryotes (Charoensawan et al. 2010). One is the prevalence of intrinsically disordered regions which are particularly associated with transcription factors (Liu et al. 2006; Singh and Dash 2007; Minezaki et al. 2006). Another eukaryotic feature that is receiving strong intense attention is the phase-separating properties of eukaryotic proteins that coalesce to form membrane-less nuclear organelles (Zhu and Brangwynne 2015; Wheeler and Hyman 2018). Liquid-liquid phase separation by globular proteins such as lysozyme has been known for some time (Ishimoto and Tanaka 1977), and this process is favored when disordered regions are involved (Uversky 2017). The third feature is the combinatorial control of eukaryotic transcription factors. Combinatorial control refers to the formation of functionally distinct factor/DNA complexes as directed by DNA sequence and protein-protein partnerships (Remenyi et al. 2004; Pan et al. 2009, 2010). These complexes, which vary discontinuously in structure and function with binding affinity, are postulated to arise from an allosteric relationship between DNA on the one hand and co-regulatory protein partners on the other. The re-purposing of eukaryotic factors through combinatorial control is central to their ability to regulate complex genomes using a restricted set of DNA-binding motifs (Vaquerizas et al. 2009; Charoensawan et al. 2010).

The profound differences between prokaryotic and eukaryotic gene regulation suggest that principles derived from bacterial and bacteriophage regulators are insufficient to understand functionally important properties of eukaryotic transcription factors. New experimental tools, notably high-throughput nucleic acid sequencing and array approaches, have revealed many novel properties of transcription factors. These results have not been always interpreted according to the combinatorial control paradigm in connection with the immediate research questions which they address. In this review, we will first review the basic tenets of combinatorial control. We will then assess recent literature on novel modes of genomic control by transcription factors, specifically low-affinity recognition. Integrating this new knowledge, we will propose a more general abstraction of combinatorial control that could help frame incisive new directions of research in this area.

2. Chapter 1. DNA Sequence and Protein-Protein Partnerships Mutually Direct Transcription Factor Interactions

In 1998, Lefstin and Yamamoto proposed that distinct DNA sequences direct conformational changes that specify interactions with other protein partners (Lefstin and Yamamoto 1998). In the context of the major transcription factor families that have been studied up to that time, such as the POU family, nuclear receptors, and AP-1, most of which recognize DNA as oligomers, Lefstin and Yamamoto drew attention to the ordering of quaternary structure by DNA sequence. Once formed, these DNA-directed complexes present alternate surfaces to recruit specific co-regulatory partners and elicit distinct transcriptional outputs. Respecting the adage that “structure equals function”, the disjoint nature of eukaryotic transcriptional responses implies significant differences in structure as they are tuned from eliciting one pattern of transcriptional behavior to another. A prescient observation by Lefstin and Yamamoto is that allosteric control by DNA is engrained in the DNA-binding domains (DBDs) and not strictly dependent on the transactivation or other domains of transcription factors.

Following the proposal by Lefstin and Yamamoto, many studies have found that subtle changes to factor-specific binding sites in regulatory regions of genes resulted in gross changes in gene expression patterns. For example, the cell type-specific regulation of growth hormone (GH) expression by the POU-family factor Pit-1 is determined by the sequence of a single Pit-1 recognition site in the GH promoter (Scully et al. 2000). Substitution of this sequence alone with an affinity-neutral counterpart abolishes the normal selectivity of GH expression in somatotropic cells. Similarly, single nucleotide changes in the cognate sequences for NF-κB, which binds DNA as dimers, alters its specificity for co-factors (Leung et al. 2004). Since the altered sequences do not differ significantly in terms the affinity of factor-specific binding in these examples, the divergence in gene expression is remarkable.

High-resolution details on how DNA sequence initiates divergent transcriptional effects come from co-crystallographic studies of several transcription factors that bind DNA in distinct modes as directed by sequence. One of the more dramatic examples is afforded by the POU-family transcription factors. Their DNA-binding (POU) domain is a bipartite combination of two helix-turn-helix motifs. POU consists of a globular POU-specific subdomain (POU_S) connected via a flexible variable-length linker to a homeodomain (POU_H) (Sturm and Herr 1988). Depending on the spacing of the POU_S− and POU_H-binding sequences in native POU cognate sites, which do not differ significantly in affinity, POU domains dimerize with the two sub-domains of each subunit arranged in different orientations along the response elements (Remenyi et al. 2001; Herr and Cleary 1995; Jacobson et al. 1997). These orientations, in which the intra-subunit sub-domains are roughly cis or trans with respect to the major groove, expose altogether different surfaces to each other as well as co-regulators (Fig. 2). Variation of the cognate binding site for the glucocorticoid receptor (GR), a nuclear hormone receptor, results in little correlation with binding affinity in vitro, but recruits different protein partners and different expression levels of GR-dependent genes (Meijsing et al. 2009; Thomas-Chollier et al. 2013). More recently, studies on the dimeric T-box transcription factor T-bet showed that subunits bound to core motifs located on different DNA duplexes can crosslink in a configuration that mobilizes more than five-fold more exposed surfaces than canonical dimers binding minimally spaced cores on the same duplex (Liu et al. 2016a). Thus, quaternary structure represents an efficient handle for allosteric control by DNA. Functionally, observation of these effects in factors belonging to unrelated structure classes (as exemplified by the IFN-β enhanceosome, Fig. 1) and in species ranging from yeast to humans suggested them as a paradigm of eukaryotes gene regulation. Not surprisingly, the factor-directing properties by DNA sequence (Jolma et al. 2015) have encouraged interest in deciphering the impact of spacing, density, and orientation of transcription factor binding sites, or “syntax”, on the cis-regulatory potential of promoters and enhancers (Baetu 2011; Barolo 2016; Witzany 2017).

Fig. 2. DNA-directed structures formed by POU-domain factor Oct-1.

Two characteristic cognate response elements, termed PORE and MORE, are bound with similar affinities. On the longer PORE site, POU_S and POU_H straddle the helix, while they sit on the same side on the shorter MORE site and induce van der Waals contacts on POU_S (yellow). Arrows show the connectivity of the unresolved linker in the co-crystal structures (PDB IDs in brackets)

Allosteric control by DNA represents one lever of the combinatorial tug-of-war for transcription factor response. Co-regulating protein partners represent the other. To cite one dramatic example, a splice variant of GR harboring one additional Arg residue in the DBD, which does not alter binding affinity, differentially activates a subset of GR-dependent genes (Meijsing et al. 2009). The definition of combinatorial control under its current moniker formally acknowledges the reciprocal influences of DNA sequence and protein partner identity (Remenyi et al. 2004). A subsequent formulation by Nussinov and collaborators in 2009 sharpened this picture by framing transcription factor interactions in terms of the free energy landscape (Pan et al. 2009, 2010). Free energy landscapes are thermodynamic constructs that relate the relative free energies of a chemical species differing in terms of conformational change or binding. Thus, a species is energetically favored relative to all others and at equilibrium, the lowest-energy state is the most highly populated in the ensemble. The energy barriers between related states also provide information on the kinetics of interconversion. While quantitative resolution of free energy landscapes of interactions between macromolecules remains extraordinary in computational cost (Pan et al. 2019), and is not routinely affordable to most investigators, they are intuitive constructs for modeling the relative (thermodynamic) stabilities of non-covalently linked states.

Based on the Nussinov concept (Pan et al. 2009, 2010), Fig. 3 shows free energy landscapes for hypothetical transitions by an ensemble consisting of three distinct DNA sequences encoding binding sites for a (homodimeric) transcription factor, and two partner proteins that are recruited via strictly protein-protein interactions. Since DNA sequence represents the only variation among the ensembles, the scheme demonstrates how specific binding remodels the free energy landscapes such that different factor/DNA complexes become the thermodynamically favored species. To the extent that these species are sufficient in mobilizing the regulatory machinery, such schemes illustrate how combinatorial control may operate in thermodynamically consistent terms. Although the various bound states transition in Fig. 3 in a particular order (beginning with DNA binding), the relative stabilities of the states are not order-dependent as per thermodynamics. The barriers separating any two adjacent states would of course apply to that specifically ordered pair. In the sense that the free energy landscape provides for cooperative control equally from protein or DNA, it captures and generalizes the DNA-directed proposal by Lefstin and Yamamoto (Lefstin and Yamamoto 1998). It is also compatible with transcription regulation at the higher level of the nucleosome, namely the reciprocal effects of nucleosomal dynamics on transcription factor binding (He et al. 2010; Grossman et al. 2018; Ballare et al. 2013), particularly by so-called pioneer factors (Magnani et al. 2011; Soufi et al. 2015).

Fig. 3. Schematic free energy landscapes for combinatorial docking to three different cognate DNA sequences.

The black-and-white DNA cartoon is intended to represent any one of the three sequences. Sequence-specific binding by a homodimeric transcription factor with affinities which are not necessarily identical (as the case here) differentially remodels the free energy landscapes (presented in black, blue, and red) for the recruitment of other proteins. The resultant changes in the most favored functionally competent complex (denoted with correspondingly colored backgrounds) bias the transcriptional output towards the pattern regulated by that complex. Even in this illustrative example, many alternative landscapes featuring other possibilities, for example co-regulating partners that recognizes one complex but not another, are possible

The free energy framework has some limitations. First, while the DNA and protein partners are not treated differently in a simplified free energy landscape such as Fig. 3, features of the genomic environment (e.g., the embedding of the target DNA site between nonspecific DNA in topographically complex chromatin, other DNA-binding proteins) that are not included may introduce asymmetry into this tug-of-war. A more general limitation is that the free energy landscape assumes that conditions of thermodynamic equilibrium prevail, at least locally, such that the relative population among the possible states are determined by their relative free energies. This assumption also requires that all accessible states be interconvertible (a consequence of the principle of microscopic reversibility), which may be slow relative to physiologically relevant timescales (Xhani et al. 2020). Nevertheless, the free energy landscape provides a useful framework for abstracting molecular details provided that all the species are known and accurately included in thermodynamic terms. One may rationalize, for example, why the affinities of factor/DNA interactions have not evolved to be typically much closer to the upper limit for non-covalent interactions (~10⁻¹⁵ M in dissociation constant), as the subsequent landscape remodeling would render energetically unfavorable any high-order state with additional partners.

As is the case of all thermodynamic constructs, free energy landscapes provide no explicit structural information on the transition. The microscopic structures or dynamics responsible for allostery become thermodynamically manifest only through their effects on macroscopic experimental observables. Dissection of the free energy changes into more fundamental components can provide useful clues based on the changes in characteristic physicochemical properties of the transition. Changes in enthalpy, entropy, and heat capacity provide information (and in the best cases, yield semi-quantitative estimates) of the amount and type of surface area involved in a transition (Spolar and Record 1994). Perturbation by hydrostatic or osmotic pressure can be modeled to give complementary information about the change in dynamics and molecular hydration (Silva et al. 2001; Royer and Winter 2009). While the analytical methodology of these types of thermodynamic analyses are quite well characterized, their limitations understood, and have been widely reported in the literature for well-defined transitions involving model domains, their use in multi-component complexes that typify functionally interesting transcription factor/DNA species remain virtually unknown.

3. Chapter 2. Low-Affinity Transcription Regulation: New Perspectives to Combinatorial Control

Recent advances in high-throughput binding assays adapting next-generation sequencing and microarray technologies have significantly improved the definition of genomic occupancy by transcription factors in vivo (Rastogi et al. 2018). These technical refinements have provided access to the population of factor/DNA complexes in vivo that are significantly lower in affinity. These studies have revealed an unanticipated high level of low-affinity binding by eukaryotic factors (Wang et al. 2015; Crocker et al. 2016; Kribelbauer et al. 2019). More importantly, mounting evidence shows the absolute requirement of low-affinity binding for correct tissue development in cellular and animal models (Farley et al. 2015; Farley et al. 2016; Crocker et al. 2015; Cary et al. 2017), as well as the exclusivity of low-affinity complexes to specific cellular functions (Krishnamoorthy et al. 2017; Iwata et al. 2017). A major area of interest in low-affinity transcriptional control in recent years has been embryogenesis. Functional studies in model animals including Ciona sea squirts (Farley et al. 2015; Farley et al. 2016), Drosophila spp. (Jiang and Levine 1993; Crocker et al. 2015; Ramos and Barolo 2013), C. elegans (Gaudet and Mango 2002), sea stars and sea urchins (Cary et al. 2017), and the annelid P. dumerilii (Handberg-Thorsager et al. 2018) show that low-affinity binding by transcription factors is essential to correct spatiotemporal patterning of tissues. Low-affinity binding is thought to enable spatio-temporal control of genes by tuning the sensitivity of their enhancers to transcription factor levels as well as facilitating cooperative regulation by multiple factors (Crocker et al. 2016; Weingarten-Gabbay and Segal 2014). Substituting the wildtype low-affinity sites with high-affinity sequences results in ectopic and overly intense gene expression, demonstrating the functional significance of low-affinity signaling in regulating the level, location, timing, and specificity of developmentally sensitive genes.

The key question for low-affinity binding is how it can deliver functionally specific transcription factor complexes. Investigators have addressed this issue in their recent reviews on low-affinity site recognition (Crocker et al. 2016; Kribelbauer et al. 2019). Proposed mechanisms of specificities involving low-affinity binding fall broadly into two categories. The first consists of mechanisms that directly generate specificity through cooperative binding, either intramolecularly or intermolecularly, to unveil latent specificities. Intramolecular avidity can be derived from combining multiple DBDs from the same or different structural families within the same protein, with zinc finger proteins such as the CTCF insulator being dramatic examples (Lambert et al. 2018). Alternatively, multiple factors assemble as dictated by DNA sequence. Hematopoiesis, an established paradigm of transcriptional control (Orkin and Zon 2008), provides important examples. The hematopoietic master transcription factor PU.1, a member of the ETS transcription factor family, exhibits at least two distinct genomic binding modes in vivo. One type is cooperative binding with partners such as IRF4 at composite high-affinity sites (Mohaghegh et al. 2019; Hosokawa et al. 2018). In cell lineages in which PU.1 expression is silenced, such as most T cells, Ap-1 takes the place of PU.1 (Glasmacher et al. 2012), as exemplified by the IFN-β enhanceosome (Fig. 1). The second type of genomic PU.1 binding is collaborative binding with factors such as C/EBPα which bind at separate sites and is triggered at low-affinity PU.1 motifs. While high-resolution differences between high- and low-affinity bound PU.1 are not yet known, biophysical evidence supports significant differences in the disposition of solvent-exposed surface areas and their distinctiveness from other members of the ETS family (Wang et al. 2014; Xhani et al. 2017; Albrecht et al. 2018).

Alternatively, to overcome the intrinsic deficit in affinity, mechanisms are proposed to modulate the specificity of transcription factors indirectly through their local concentrations. Eukaryotic subcellular compartments, dubbed transcriptional hubs, are postulated to sequester transcription factors (Kribelbauer et al. 2019). Liquid-liquid phase separation (LLPS) into membrane-less organelles is thought to be one hub-generating mechanism (Zhu and Brangwynne 2015; Wheeler and Hyman 2018). Notwithstanding the interest in disordered proteins with LLPS, due to their abundance in nuclear proteins (Uversky 2017), structured globular proteins also form similar condensates under appropriate solution conditions (Cinar et al. 2019). It should be noted that, to be a useful pool, the sequestered transcription factor must be physically available for exchange with genomic targets. The DBD of GR, for example, binds with structured RNA encoded by different sequences with comparable affinity as canonical DNA response elements (Parsonnet et al. 2019). Since RNA-bound proteins are now widely known to undergo LLPS (Buchan 2014; Rhine et al. 2020), and GR could hardly be alone among transcription factors to bind RNA, sorting out such sequestration mechanisms in support of low-affinity transcriptional regulation represents a complicated proposition.

The widespread existence and functional importance of low-affinity binding motivate the need to interpret it in the context of the combinatorial control framework, which envisions sites selection among DNA sites of similar affinities in the free energy landscape (Pan et al. 2009, 2010). As long as some of the DNA-bound states are more stable than the unbound transcription factor, binding redistributes the bound and free states of the transcription factor with additional potential of remodeling the free energy landscape. Thus, allosteric control by low-affinity DNA does not necessarily present any contradiction. However, if one assumes stationary conditions at thermodynamic equilibrium, population of the various states is entirely dependent on concentration according to the law of mass action. Lower-affinity interactions therefore demand large concentrations of unbound components, particularly in the presence of higher-affinity competitors. This may mean concentration requirements for low-affinity targets that are nominally non-physiological. Living cells, of course, are not under thermodynamic equilibrium overall (though chemical potentials may be near their equilibrium values locally), and various sequestration mechanisms have been proposed that may significantly localize and boost effective concentrations (vide supra). Moreover, transcriptional effects only require sufficient kinetic persistence at the level of individual molecules, which reflects but is not dictated by equilibrium properties of macroscopic ensembles.

In addition to thermodynamic concerns, low-affinity transcriptional complexes pose structural challenges to the combinatorial control framework. A striking example is afforded by an alternative dimeric binding mode by GR to DNA motifs in which a high- and low-affinity core are oriented tail-to-tail (an “inverted repeat”) such that the subunits lie on opposite faces of the DNA (Weikum et al. 2017). Unlike the well-known canonical motif in which two high-affinity cores recruit the subunits avidly with positive cooperativity in a spaced head-to-head configuration (Luisi et al. 1991), inverted repeats are bound 50–100 times more poorly and with negative cooperativity (Hudson et al. 2013). The co-crystal structure shows a that the GR monomers engage each core as separated monomers in the inverted repeat, which characterizes repressive GR-responsive elements, in stark contrast to the cooperative dimers found in the canonical complex bound to activating response elements (Fig. 4). GR therefore demonstrates that low-affinity complexes can be structurally so different that their free energy landscapes would be altered quite beyond the fine-tuning that characterizes combinatorial control (c.f., Fig. 3) as originally envisioned for cognate sequences of comparable affinities.

Fig. 4. Alternate bound states of the glucocorticoid receptor (GR) DBD dimer.

The canonical GR/DNA complex (red; PDB ID: 1glu) and an alternate tail-to-tail orientation (blue; PDB ID: 4hn5) are structurally aligned at the middle monomer to highlight the inverted orientations (arrows). Zn²⁺ ions coordinated by the two zinc fingers in each monomer are shown as spheres. The hexamer core sequences recognized by the monomers are marked in grey. Unlike the canonical dimer, the latter orientation is characteristically low-affinity, exposes different surfaces to each monomer and co-regulators, and mediates trans-repressive functions of GR

4. Chapter 3. Rethinking Combinatorial Control: Concept of Non-continuum Response

In the combinatorial control framework, transcription factors are flexible species that can be allosterically repurposed by DNA and protein in an identity-dependent manner. The mechanisms of allosteric regulation are myriad. Investigations on the DNA side alone have revealed profound effect of short-range DNA sequence and longer-range cis-regulatory “syntax” (spacing, density, orientation) that extends well beyond the sequence of the transcription factor binding site itself. More dramatic though less common are locally non-canonical DNA structures, such as triplex and G-quadruplexes, which can directly bind (Cogoi et al. 2010; Gao et al. 2015) or modify recognition by transcription factors (Zhou et al. 2019). On the protein side, quaternary structure and heterotypic protein-protein partnerships can amplify permutations in binding both locally and with distal cis elements potentially hundreds of bp away in sequence. This flexibility in the structure as well as usage of transcription factors and their DNA binding sites has been evocatively termed “plasticity” and “pleiotropy” (e.g., Kyrmizi et al. 2006; Weikum et al. 2017; Preger-Ben Noon et al. 2018; Kalodimos et al. 2002). Readers may recognize these borrowed metaphors from the description of complex organismal and cellular phenotypes in the genetics and developmental biology literature.

To incorporate low-affinity regulation more seamlessly into combinatorial control, and secondarily to avoid the potentially contentious usage of domain-specific labels, we propose a conceptual framework in which transcription factors generate qualitatively disjoint, or non-continuum responses as specified by intermolecular interactions with DNA and protein partners. These partner-specific interactions direct the factors to broadcast the effects of specific transcriptional programs. Conceptually, the designation of “non-continuum” focuses attention on the effects of transcription factor action, rather than the manner in which they are achieved. As a result, it conveniently unites combinatorial control of transcription factors via at high- and low-affinity DNA sites, which have been alternately described as “suboptimal” or “submaximal” (Farley et al. 2015; Farley et al. 2016). Finally, it strikes an important contrast with the prokaryotic paradigm of transcriptional regulation.

Prokaryotic gene regulators represent some of the best understood systems in transcriptional regulation. Systems such as the cAMP-response protein (CRP, also known as catabolite activator protein or CAP) and the lac repressor are classic models in our understanding of the operon and protein/DNA allostery (Kalodimos et al. 2004b; Lawson et al. 2004). A general property in prokaryotic gene regulation is that specific regulators have narrow functional scopes, meaning that they mediate consistent functions at their DNA target sites, even though they may be combined in complex cis-regulatory logic (van Hijum et al. 2009). This description recognizes that control regions of prokaryotic operons may be expansive in organization, as exemplified by the multi-operator architecture of the lac and trp operons (Merino et al. 2008). Prokaryotic proteins also exhibit complex allostery that coordinates the mutually perturbative effects of binding to DNA and low-MW metabolites (Popovych et al. 2006). However, the activity and molecular interactions of these proteins track each other quantitatively along a gradient, generally as a function of their affinities for the DNA site or other ligands. Detailed NMR-based studies of CAP/DNA binding, for example, show a continuum transition in structure in step with affinity, as indicated by the chemical shifts of heteronuclear cross-peaks (Tzeng and Kalodimos 2012). Indeed, this continuum between structure and affinity makes possible the exhaustive characterization of DNA sequence by high-throughput screening of operator libraries (Zuo et al. 2015; Zuo and Stormo 2014).

With this backdrop, “non-continuum responses” denote the qualitatively distinct molecular behavior of eukaryotic transcription factors that are elicited upon binding different cognate DNA sequences and protein co-regulators (Fig. 5). Unlike gene regulation in prokaryotes, these outcomes in gene expression and interactions with partner proteins cannot be graded according to binding affinity with any one partner. As GR and others systems exemplify, eukaryotic transcriptional responses triggered by low-affinity binding are therefore not just fractional in magnitude relative to high-affinity binding. Instead, they are qualitatively non-equivalent and biologically irreplaceable as reviewed in Chapter 2. Thus, for both high- and low-affinity binding, “non-continuum responses” capture the essential molecular tug-of-war between DNA and protein co-regulators that remodels the conformations of transcription factors discontinuously in structure.

Fig. 5. Non-continuum nature of output marks eukaryotic transcription factors.

In prokaryotes, gene regulatory proteins elicit qualitatively consistent functions (e.g., repression of an operon) whose intensity quantitatively tracks the DNA binding affinity. Continuum behavior can be experimentally observed, for example, by solution NMR via heteronuclear chemical shifts e.g., (Tzeng and Kalodimos 2012). In stark contrast, eukaryotic transcription factors elicit qualitatively disjoint effects (e.g., activity in a particular cell type, collaboration with specific co-regulators) that do not correlate with binding affinities

5. Chapter 4. Dynamic Mechanisms in Non-continuum Responses

5.1. Section 4.1. Importance of Conformational Dynamics in Combinatorial Control

Biopolymers in general, including proteins and nucleic acids, are conformationally dynamic objects in solution. Dynamics arise most fundamentally from significant thermal motion at temperatures at which water is liquid. Globular protein domains and helical DNA (and folded RNA) are restrained dynamically relative to their disordered chains, as quantified thermodynamically by the configurational entropy. Nevertheless, significant dynamics remain in the folded states of proteins, as has been realized since at least the 1970s (Cooper 1976, 1984). In eukaryotic proteins, these dynamics can be concentrated locally in intrinsically disordered regions (IDRs). Compared with prokaryotic gene regulators, transcription factors are particularly enriched in intrinsic disorder, particularly near their structured DNA-binding domains (Guo et al. 2012). Among a host of other molecular functions, IDRs are also frequent targets of post-translational modifications (van der Lee et al. 2014). From yeasts to high-order eukaryotes, transcription factors therefore appear to accrue increasing dynamics in their natural molecular evolution, suggesting physiological significance.

Given the fundamental dynamics of proteins and nucleic acids, it is natural to consider how conformational dynamics of transcription factors mediate their responsiveness to DNA targets and partner proteins in an identity-dependent manner. These considerations dovetail to long-standing questions of conformational selection (capture of pre-existing conformations within the ensemble of the free state) versus induced fit (conformational adjustment), a classic debate with origin in the concerted and sequential models of allosteric transitions (Monod et al. 1965; Rubin and Changeux 1966). The abundance of intrinsic disorder in transcription factor adds to the functional relevance of these lines of inquiry. Following up on their work on combinatorial control, Nussinov and collaborators postulated that these dynamics primarily involve conformational selection by DNA or partner protein, potentially with induced fit along the binding pathway (Nussinov et al. 2014). Although this picture is in accord with contemporary consensus derived from hemoglobin, model enzymes and membrane receptors (Vogt et al. 2014; Vogt and Di Cera 2012, 2013; Changeux and Edelstein 2011), it is nevertheless inferred from static structures (Nussinov et al. 2014; Pan et al. 2010).

Commenting primarily on GR, Leftsin and Yamamoto had proposed that sequence-dependent signals are transduced from the protein/DNA interface to allosteric sites postulated to recruit protein co-regulators (Lefstin and Yamamoto 1998). To test such hypotheses, it would be necessary to define the dynamics of the unbound protein and the dynamic perturbations due to binding. The implied dynamic perturbations by high- and low-affinity DNA would be expected to differ substantially, with distinct structural and thermodynamic signatures. Unfortunately, direct investigations of transcription factor dynamics have received less attention relative to other classes of proteins. Moreover, this knowledge gap is not well addressed by the significant work done on prokaryotic regulators (Popovych et al. 2006; Villa et al. 2005; Kalodimos et al. 2004a), which act as continuum transcriptional modulators.

In the literature, the non-continuum dynamics of two transcription factor systems have been examined in some detail. The first is GR, whose DBD engages DNA via two Cys₄ zinc fingers to cognate hexameric core sequences. The Cys₄ zinc fingers are separated by a loop termed the “lever arm” which is not canonical to this common DNA-binding motif (Vandevyver et al. 2014). Comparison of co-crystal structures show that the lever arm is conformationally variable and modifies the relative orientation of the two zinc fingers discretely in a sequence-dependent manner (Meijsing et al. 2009). Lever arm dynamics have been examined directly by solution NMR spectroscopy (Thomas-Chollier et al. 2013) and in silico by explicit-solvent molecular dynamics (MD) simulations (Alvarez et al. 2017). The results show that lever arm dynamics are sensitive to clinically relevant point alterations in primary structure, and are propagated to distal interfaces that mediate GR dimer formation and interactions with co-regulators.

A second model system is represented by Ets-1, the archetypal member of the ETS family of transcription factors. ETS transcription factors comprise an ancient family of gene regulators and are conserved throughout the animal kingdom (Degnan et al. 1993; Wang and Zhang 2009). ETS proteins recognize diverse DNA sequences harboring a core 5′-GGA(A/T)-3′ consensus (Wei et al. 2010). Among many other roles in cellular housekeeping and hematopoiesis, Ets-1 is specifically implicated in correct development of the notochord in Ciona embryos (Farley et al. 2015; Farley et al. 2016). A low-affinity site in the Mnx enhancer is a specific and absolute requirement for correctly patterned Mnx expression in vivo (Farley et al. 2016). Ets-1 binds core DNA sites characteristically as monomers (Garvie et al. 2002). To resolve the structural and dynamic basis of low-affinity recognition, extensive explicit-solvent MD simulations were carried out for Ets-1 free and bound to cognate and nonspecific DNA (Huang et al. 2019). Principal component analysis (PCA) of the trajectories shows that high- (HA) and low-affinity (LA) cognate binding represent disjointed DNA recognition modes (Fig. 6a). While backbone dynamics separate cognate (HA, LA) from nonspecific (NS) DNA recognition, sidechain dynamics distinguish HA and LA binding. By both back-bone and sidechain dynamics, the HA complex is more similar to nonspecific than LA (black arrows, Fig. 6a). Thus, LA-bound Ets-1 is a non-continuum species residing outside the unbound-NS-HA coordinate.

Fig. 6. Non-continuum Ets-1/DNA recognition.

(a) Principal component analysis of MD trajectories of the ETS domain of Ets-1, free and in complex with DNA, reveals collective dynamics out of order with their binding affinities. Red arrows denote the collective sidechain motions along the first principal component, PC1. HA = high-, LA = low-affinity, NS = nonspecific, UB = unbound. (b) A triad of residues acts as a binary relay. LA binding breaks an allosteric salt bridge that is constitutively intact in the other states. (c) Disruption of this salt bridge floods an underlying hydrophobic core in helices H1 and H2. (d) DNA binding by a Q336L mutant, which renders the relay unresponsive in simulations, fails to trigger low-affinity binding. Data published in (Huang et al. 2019)

Scrutiny of the sidechain PCA and inter-residue contacts reveals a triad of residues that mediate the non-continuum LA response (Fig. 6b). Gln³³⁶ and Arg³⁷⁸ near the DNA-contact surface, together with a distal Glu³⁴³ residue, act as a binary relay that toggles between two sidechain configurations. In the default configuration (as present in unbound Ets-1 and preserved in nonspecific and HA-bound complexes), Glu³⁴³ and Arg³⁷⁸ form a salt bridge, while Gln³³⁶ contacts the DNA backbone. LA DNA uniquely triggers a phosphate contact with Arg³⁷⁸ that breaks the salt bridge with Glu³⁴³. Loss of this salt bridge destabilizes the local α-helical structure and exposes an underlying hydrophobic patch to solvent (Fig. 6c). Thus, cognate DNA recognition by Ets-1 involves an allosteric mechanism in which sequence variations are sensed by an interfacial Arg residue and transmitted to a distal salt bridge. Abrogation of this salt bridge exposes and hydrates the hydrophobic core, incurring a binding penalty. In simulations, a Q336L mutation desensitizes the relay to LA DNA. Experimentally, Q336L binds LA DNA with similarly high affinity as HA DNA, thus failing to trigger the LA-specific response (Fig. 6d). Q336L also fails to reproduce the site occupancy pattern of wildtype Ets-1 in a native enhancer context (Huang et al. 2019). Binding experiments with Ets-1 mutants therefore verified the functional effect of the triadic relay. Conservation of the triadic relay in ~50% of ETS members suggests that it may be a shared non-continuum mechanism in this family.

Though still in limited supply, these examples of transcription factors, from unrelated structural families and whose non-continuum transcriptional properties are functionally established, illustrate a diversity in their dynamic mechanisms of DNA recognition. These examples suggest a general theme in which DNA-binding motifs evolve novel mechanisms to overcome structural constraints to broaden their DNA sequence tolerance. The attendant evolutionary pressure offers one resolution to the general observation that “no simple code” exists in protein/DNA readout (Slattery et al. 2014) when average shape, but not dynamic properties, is explicitly taken into account.

5.2. Chapter 4.2. Molecular Hydration in Non-continuum Tuning

The nucleoplasm is a crowded microenvironment in which a complex mixture of solutes ranging over orders of magnitude in size, concentration, and physicochemical properties populate an aqueous milieu (Goodsell 1991). As the space-filling solvent, water remains the numerically dominant species, and its overwhelming demand for hydrogen bonding (Eaves et al. 2005) strongly influences biomolecular conformation and binding. Hydration is therefore an indispensable component in protein and nucleic acid dynamics in solution (Chalikian 2003; Chalikian and Macgregor 2007), including in vivo (Parsegian and Rau 1984; Rand et al. 2000). The close connection between hydration and dynamics is a classical topic in physical organic chemistry and continue to garner strong interest for their roles in biomolecular interactions. Their well-known thermodynamic coupling, enthalpy-entropy compensation, remains actively debated (Fox et al. 2018; Pan et al. 2015). In the literature, work has focused on the hydration dynamics of folded and unfolded states of proteins and nucleic acids (Duboue-Dijon et al. 2016; Laage et al. 2017; Persson et al. 2018). The results have updated our view of hydration water as highly heterogeneous and generally more mobile than previously considered. Work on biomolecular recognition has focused on protein binding by small ligands e.g., (Liu et al. 2016b; Breiten et al. 2013). The coupling of hydration and conformational dynamics in DNA recognition remains an uncultivated area despite obvious differences between DNA and low-MW substrates. As a polyanion, helical DNA has significant sequence-dependent dynamics and hydration that can impact binding. Addressing these fundamental concepts in sequence-directed tuning should expand our knowledge of protein/DNA recognition beyond time-averaged sequence-specific DNA structures and shape (Rohs et al. 2010; Joyce et al. 2015). The triadic relay in Ets-1, which works directly through hydration of a hydrophobic core in the low-affinity complex (Huang et al. 2019), represents one such example.

5.3. Chapter 4.3. Experimental and Computational Approaches to Molecular Dynamics and Hydration

X-ray crystallography is the classic high-resolution probe of macromolecular structure and provides dynamics information indirectly through the temperature factor and multiple occupancies of local electron densities (Wlodawer et al. 2008). Cryo-electron microscopy is now approaching, in the best cases, resolution of crystallography for large assemblages with dimensions in the mesoscopic length scale (10–1000 nm). The information content of Cryo-EM, including dynamics information, is extended still when structure fitting occurs in concert with crystallographic structures of components and MD methods. For true in-solution dynamics, high-field multi-dimensional NMR is a powerful technique for probing the dynamics of proteins and nucleic acids (Kleckner and Foster 2011; Purslow et al. 2020). As NMR measures the macroscopic ensemble, it benefits greatly from atomistic modeling of dynamical observables by MD approaches. MD is also highly complementary to single-molecule experiments provide direct observation of dynamics at the mesoscopic length scale (Deniz et al. 2008).

With respect to hydration, water molecules are frequently observed in crystal structures. However, discrete water observed in crystal structures constitutes the most stable, well-ordered sub-population of the water involved in hydration (Nakasako 2004). The more mobile water associated with dynamic structures do not produce observable electron densities. NMR is capable of probing hydration of macromolecular surfaces in solution, but reliably only if the macromolecules are encapsulated in reverse phase micelles to break the magnetic coupling between local and bulk water (Gallo et al. 2018). These complications have thus far limited the application of NMR in hydration studies. By far the most frequently used approach to hydration of nucleoprotein complexes is osmotic stress, using neutral compatible co-solutes such as betaine and sucrose to perturb hydration water via an increase in osmotic pressure in bulk solution (Parsegian et al. 1995). Another volumetric technique that has been applied in protein/DNA binding is hydrostatic pressure, though much more commonly for enzymes than for transcription factors (Macgregor 2002). A vast literature exists on the volumetric properties of the folded and unfolded states of model proteins, polypeptides, nucleic acids, and their interactions with low-MW ligands; reviewed e.g., (Chalikian 2003; Chalikian and Macgregor 2007; Royer and Winter 2009). As suggested by Sligar (Lynch and Sligar 2002), modeling of volumetric data with MD simulations have the potential to yield significant insight. In addition to conformational dynamics of macromolecules, MD simulations have also been used to analyze the dynamics of hydration water purely in silico (Laage and Hynes 2006; Duboue-Dijon et al. 2016; Persson et al. 2018).

6. Chapter 5. Evolutionary Origins of Non-continuum Tuning

How did non-continuum behavior arise in eukaryotic transcription factors? Some evolutionarily modern transcription factors harbor de novo DNA-binding motifs e.g., T-box proteins (Sebe-Pedros et al. 2013), or heavily modified classical motifs such as nuclear receptors (de Mendoza et al. 2013; Schmitz et al. 2016). However, many factors harbor DNA-binding motifs that are directly descended from prokaryotic gene regulators. Since the molecular properties that endow non-continuum transcriptional outputs are already engrained in the DBDs, this suggests an evolutionary relationship between continuum and non-continuum behavior in the most conserved motifs across the trees of life. The helix-turn-helix (HTH) domain is an ideal model in this regard, because it is one of the most ancient and conserved of the transcriptional apparatus in all organisms (Sauer et al. 1982) and was already present in the last universal common ancestor (LUCA) of extant life forms (Burton et al. 2016). HTH motifs are extraordinarily robust to sequence changes as evidenced by the myriad amino acid sequences capable of encoding them (Aravind et al. 2005). As examples, we discuss two lineages of HTH descendants: cAMP-response protein (CRP)-like motif and tetrahelical HTH bundles characterized by bacteriophage proteins.

6.1. Chapter 5.1. CRP-Like Motif

The CRP-like motif is a mixed α/β bundle in which a 4-stranded, anti-parallel β-sheet is interrupted by an HTH core in the primary structure (Aravind et al. 2005). The ETS, interferon regulatory factors (IRF), and heat shock factor (HSF) families represent the extant eukaryotic descendants of the CRP-like motif (Liang et al. 1994; Landsman and Wolffe 1995). In bacteria, the DBD of CRP is integrated with its N-terminal cAMP-sensor domain in a one-component system. Two CRP subunits bind DNA as a homodimer that is connected via the sensor domains only. The stimulatory effects of cAMP on DNA binding in CRP, as well as the cooperativity in cAMP binding across the two CRP subunits, represent a classic model of protein allostery (Lawson et al. 2004). The tertiary structure of the DBDs in CRP is strongly constrained by the encroaching sensor domain (Fig. 7a). Detailed NMR characterization has firmly established the continuum nature of the conformational CRP trajectory from unbound to the cognate complex (Tzeng and Kalodimos 2012). In contrast, all three eukaryotic families exhibit with non-continuum properties with low-affinity DNA (Santoro et al. 1998; Iwata et al. 2017; Krishnamoorthy et al. 2017). Beyond their divergent primary structures, the eukaryotic CRP-motifs are differentiated from CRP by their elaboration of structural elements, particularly loops (Fig. 7b). These elaborations inflate the eukaryotic CRP-motifs by up to ~40% in sequence length, suggesting that inflated domain structure may represent an enabling feature for non-continuum dynamics.

Fig. 7. Evolutionary relationships in two HTH-derived motifs.

(a) Average NMR structures of the three DBD families that descended from CRP. The non-DNA-binding sensor domains in homodimeric CRP are in grey. (b) Comparison of the secondary structure elements between the eukaryotic factors and CRP. Positive values indicate elaboration in the eukaryotic factors. Green bars represent loops adjoining neighboring structured elements. Preferential elaboration of loops is evident. (c) Structural alignment of the POU_S domain with HTH bundles from the 434 bacteriophage (all NMR average structures). A conserved intramolecular salt bridge confirms and emphasizes this conservation (PDB IDs in brackets)

6.2. Chapter 5.2. POU Transcription Factors

The bipartite architecture of POU-domain factors, which has been discussed in Chapter 1, is a clear eukaryotic innovation and the linker’s diversity in length and composition has been characterized (van Leeuwen et al. 1997). The two POU sub-domains are not independent binding units, however, as the two complexes show distinct and non-overlapping POU_S/POU_H interfaces (Fig. 2). From an evolutionary perspective, homeodomains including POU_H represent a phylogenetically distinct innovation in eukaryotes, with no phylogenetic counterparts among extant prokaryotes (Aravind et al. 2005). In contrast, the POU_S domains, which are tetrahelical bundles, are highly conserved with the DBD of the bacteriophage 434 repressor and 434 CRO proteins (Fig. 7c), including a characteristic intramolecular Arg-Glu sidechain-mediated salt bridge (Assa-Munt et al. 1993). Bacteriophages are very ancient viruses, substantially predating the separation of archaea and eukaryotes from prokaryotes (Brüssow and Hendrix 2002). Moreover, binding studies show that POU_S, not POU_H, is the primary determinant for POU-specific binding (Ingraham et al. 1990). The evidence therefore indicates that the POU_S domain is the major behavior-modifying component and the relevant species for cross-kingdom evolutionary analysis.

7. Concluding Remarks

Eukaryotic transcription factors are remarkably versatile mediators of transcriptional regulation. Recent biophysical, genomic, and biological studies have considerably advanced our understanding of the molecular basis and functional relevance of this flexibility. Integrating the existing literature on combinatorial control and low-affinity transcriptional regulation, we propose in this review the concept “non-continuum responses” to unite these ideas and contrast them with prokaryotic transcriptional regulation. In addition, we emphasize the importance of conformational dynamics of the folded state and their attendant effects on molecular hydration in sensitizing transcription factors to a large range of co-regulatory protein and DNA partners. Conformational dynamics and hydration are classic concepts in molecular biophysics whose theoretical underpinnings have been extensively studied. A frontier in these areas is now the application of computational approaches to domain-scale systems and their complexes, in explicit solvent, to model experimental observables by ensemble techniques, such as NMR and volumetric measurements, as well as single-molecule tracking. We expect these developments to meaningfully inform genome-scale data in protein binding and gene expression for an increasing number of physiologically and clinically interesting targets.