TF paralogs – natural experiments in DNA binding

Abstract

Transcription factor (TF) paralogs provide unique insights into how DNA binding specificity evolves and diversifies. While paralogous TFs share conserved, highly similar DNA binding domains, they achieve distinct regulatory functions through mechanisms that are now being elucidated. This review examines how sequence variations between paralogs translate into functional diversity, including how mutations distant from the DNA interface can allosterically modulate binding specificity. We focus on competitive binding dynamics when paralogs are co-expressed and discuss emerging evidence that TFs recognize extensive repertoires of lower-affinity binding sites. Differential preferences for lower-affinity binding sites create paralog-specific binding patterns that determine TF genomic occupancy. These insights have important implications for interpreting the impact of coding and noncoding variation on TF-DNA interactions and human disease.

Introduction

Transcription is tightly regulated in response to intracellular and extracellular signals by sequence-specific binding of transcription factors (TFs) to DNA. Of the ~1,639 TFs encoded in the human genome, the vast majority of them contain DNA binding domains (DBDs) that belong to large families, such as Cys₂His₂ (C2H2)-type zinc fingers (n = 747; ~45%), homeodomains (HDs; n = 196; ~12%) and basic helix-loop-helix (bHLH; n = 108; ~6%) [1]. The large size of these TF families reflects diverse evolutionary mechanisms that have shaped the human genome, such as whole-genome duplications [2,3], block duplication [4], tandem duplication [5,6], segmental duplication [7,8], domain shuffling [9,10] and transposable element activity [11]. These mechanisms have generated extensive sets of paralogous TFs within each family, raising fundamental questions about how functional specificity is maintained and diversified among closely related TFs.

TF paralogs provide a unique window into the evolution of DNA binding specificity and transcriptional regulation. Regardless of their specific evolutionary origin, newly formed paralogous TFs typically begin with overlapping functions in the organism. However, over evolutionary time, they diverge in their spatiotemporal expression patterns, protein-protein interaction partners, and DNA binding preferences, while alternative splicing provides an additional mechanism for functional diversification [12–17]. Decades of genetic studies in mouse models have revealed complex functional relationships - redundancy, additivity and synergy - between paralogous TFs (Figure 1; extensively reviewed in [18]).

Figure 1:

After duplication, complex functional relationships, such as redundancy, additivity, and synergy, arise between paralogous TFs. Synergy indicates neo- or sub-functionalization, which can occur due to divergence in tissue expression patterns and/or protein sequence.

Functional redundancy occurs when paralogous TFs perform overlapping roles, allowing one paralog to sustain biological function in the absence of another. For example, Meis1 and Meis2 can compensate for the absence of each other, but their combined deletion leads to arrested lens development [19]. Functional additivity occurs when paralogous TFs exhibit dose-dependent effects, where the contribution of each functional allele quantitatively influences the phenotype. For example, loss of any 2 alleles of Dlx5 and Dlx6 leads to the same phenotypic consequence, while loss of three or all four alleles led to proportionally more severe craniofacial defects [20]. Functional synergy is detected when simultaneous knockouts of multiple paralogs reveal dependencies that are not apparent from single-gene knockouts, as revealed by the analysis of Hox3 paralogs, Hoxa3 and Hoxd3 [21]. While individual knockouts showed relatively minor and non-overlapping defects, double mutants demonstrated exacerbated phenotypes, where knockout of Hoxa3 enhanced defects in Hoxd3 mice, and vice versa [21]. Redundancy and additivity indicate that TF paralogs haven’t yet diverged completely in their functions. However, functional synergy often indicates that neo- or sub-functionalization has occurred, since the paralogs display distinct but interdependent roles.

Functional synergy observed for Hoxa3 and Hoxd3 can be explained by the divergence in their spatiotemporal tissue expression patterns [22]. Remarkably, despite sharing only ~50% amino acid identity, when the Hoxd3 protein was expressed from the Hoxa3 locus, it fully rescued Hoxa3-dependent phenotypes, and Hoxa3 protein expressed from the Hoxd3 locus complemented Hoxd3 defects [22]. Reciprocal swap experiments, therefore, show that the two Hox3 paralogs are functionally equivalent. Their distinct developmental roles arise primarily from differences in their expression patterns. Similarly, despite sharing only ~50% amino acid identity, the Hox1 paralogs Hoxa1 and Hoxb1 appeared to be functionally equivalent based on reciprocal swap experiments [23]. However, under selective pressure in semi-natural conditions, where treatment and control mice compete directly for resources and mates, significant fitness deficits emerged. The Hoxa1^B1 and Hoxb1^A1 swap mice produced only 77.9% and 64.4% as many offspring as control mice [24,25], respectively, showing that HoxA1 and HoxB1 proteins are not fully interchangeable. The protein sequence divergence between Hoxa1 and Hoxb1 led to functional specialization that only became apparent when subjected to evolutionary constraints. Therefore, divergence in expression patterns and protein sequence are two different, but non-mutually exclusive, mechanisms for neo- or sub-functionalization of TF paralogs.

Protein sequence divergence between TF paralogs often occurs in intrinsically disordered regions (IDRs) outside DBDs, while DBDs remain highly conserved among paralogous TFs, reflecting their fundamental role in sequence-specific DNA recognition. While DBDs have been extensively studied for their role in determining intrinsic DNA binding specificity, the role of IDRs in modulating TF-DNA interactions has gained increasing recognition in recent years [12,26]. IDRs within TFs have been shown to mediate protein-protein interactions (PPIs) that can influence overall DNA binding specificity. For example, the hexapeptide motif (L-Y/F-P-W-M-K/R) is highly conserved in Hox paralog groups 1–8 but diverges significantly in posterior groups 9–13. Mutation of the hexapeptide in Hoxb8 reduced cooperative DNA binding with Pbx TFs without altering its homeodomain-mediated DNA binding preference [27]. The shift in target specificity led to homeotic transformations that resembled loss-of-function mutations in other Hox genes, such as Hoxb7 and Hoxb9, likely by competing for their DNA binding sites [27]. Indeed, cooperative binding with Exd in Drosophila (Pbx in vertebrates) has been shown to evoke ‘latent specificities’ in Hox proteins [28]. This mechanism has been shown recently to orchestrate paralog-specific regulatory networks during the developmental patterning of leg segments [29]. Despite possessing nearly identical DBDs, the Hox paralogs Scr and Ubx achieve distinct regulatory functions through interactions with Exd in proximal, and Distal-less (Dll) in distal, leg domains [29]. This spatial partitioning of cofactor availability demonstrates how PPIs can modulate the DNA binding preferences of paralogous TFs and contribute to their functional diversification.

Divergence in DNA binding specificity within TF families

While functional genetic studies have revealed the complex relationships between TF paralogs, they raise fundamental questions about the molecular mechanisms that enable both functional overlap and specialization. Understanding how paralogous TFs can maintain redundant functions while simultaneously evolving distinct roles requires examining the molecular basis of their DNA binding activities. In contrast to IDRs, which have undergone extensive divergence over evolutionary time, DBDs are highly conserved among paralogous TFs, reflecting their fundamental role in sequence-specific DNA recognition. However, the subtle sequence variations that do exist between paralogous DBDs are often maintained across species, which suggests they may be functionally significant. Large-scale surveys have revealed that TF paralogs from the same family often display highly similar DNA binding motifs. This observation has contributed to a widespread assumption that minor differences in paralogous DBDs have little impact on DNA binding specificity. However, emerging evidence challenges this assumption, revealing that even small sequence variations can significantly influence TF-DNA interactions, leading to functional diversification among TF paralogs. Systematic analyses of the two largest TF families have provided key insights into how seemingly minor sequence differences lead to divergence in DNA binding specificities.

C2H2-type zinc fingers (ZF) occur in variable number of arrays (1–38) within proteins. With each ZF recognizing approximately a DNA triplet (3 bp) [30], there is a wide variety of DNA sequences of varying lengths that can be recognized by TFs containing C2H2-type ZFs. In an individual ZF, there are 6 positions that have been shown to determine specificity for particular DNA triplets. Recent bacterial one-hybrid (B1H) experiments testing all possible ~64 million amino acid combinations at these 6 specificity determining residues have shown that ZFs can be engineered to recognize all possible DNA triplets, and neighboring ZFs create distinct interface environments that selectively enable or constrain recognition of ZFs to specific DNA triplets [31]. Moreover, yeast C2H2 ZF paralogs with identical amino acids at the specificity determining positions still diverge in their binding preferences through non-canonical residues [32]. C2H2-type ZF arrays, therefore, function as highly malleable molecular scaffolds that can be tuned systematically to recognize virtually any DNA sequence of varying lengths.

Homeodomain proteins are another exemplary model where sequence diversity within a conserved structural framework generates extensive variation in DNA binding specificity. These ~60 amino acid domains exhibit remarkable evolutionary conservation across species while displaying substantial sequence diversity within genomes. For instance, the median number of amino acid differences between any two mouse homeodomains is 37 [33]. A comprehensive, high-resolution protein-binding microarray (PBM) analysis of 168 mouse homeodomains revealed that this sequence diversity translates into at least 33 distinct DNA binding activities with clearly different primary motifs [33]. Remarkably, even within each of these 33 major specificity groups, further subdivisions emerge when binding preferences are analyzed at higher resolution [33]. Paralogous homeodomains that share essentially identical primary motifs often display distinct preferences for lower-affinity binding sites, such as Lhx2 and Lhx4 [33]. The molecular basis for this specificity divergence involves amino acid substitutions at key recognition positions. While positions 47, 50 and 54 in the homeodomain recognition helix are important determinants of DNA sequence specificity, a recent high-throughput analysis of 92 human homeodomain variants, including rare and disease-associated alleles, revealed 14 previously unrecognized specificity-determining residues in homeodomains [34]. Five of these do not even contact DNA directly [34], indicating that regions outside the putative DNA-contact interface contribute to specificity in ways we currently cannot adequately explain. Moreover, the same amino-acid substitution introduced in different homeodomain paralogs often had distinct impacts on DNA binding [34]. This context-dependence highlights how each TF’s unique protein sequence background, including epistatic interactions among residues, influences how a given mutation will alter a TF’s DNA binding activity.

A systematic analysis of the forkhead family revealed an unexpected form of binding diversity. While most forkhead TFs bind the canonical RYAAAYA motif, some family members recognize a completely different GACGC sequence, and remarkably, certain ‘bispecific’ forkheads can bind both motifs with high specificity [35]. Crystal structures of bispecific FOXN3 bound to each of these sequences reveal that the same amino acid residues make different base-specific contacts [36]. The primary difference lies not in protein conformation but in DNA shape. The DNA bends away from the protein when bound to the alternative motif but toward the protein for the canonical motif, enabling the same DBD to accommodate different motifs [36]. This challenges traditional models where binding specificity changes require amino acid substitutions in DNA-contacting residues. Taken together, these findings challenge the widespread assumption that differences in DNA binding preferences between paralogous TFs are negligible. In fact, there is mounting evidence that DBDs are structural scaffolds that can be finely tuned to interact with DNA.

The functionally significant divergence observed in C2H2-type zinc fingers, homeodomain and forkhead families extends broadly across other major TF classes. When profiled for binding to genomic sequences in vitro, TF paralogs from the bHLH, ETS and E2F families displayed divergent DNA binding preferences despite previous reports that the paralogous TFs share identical binding motifs [37]. These differences in intrinsic binding preferences, which are not well captured by traditional motif models, illustrate how even minor protein sequence variations between paralogous TFs can fine-tune DNA recognition, enabling functional diversification despite overall motif similarity.

Competition for DNA binding between paralogous TFs

The fact that paralogous TFs can possess subtly different DNA binding preferences, despite their overall sequence similarity, has important implications for gene regulation. Co-expression of paralogous TFs is quite common in many cell types and developmental stages [38]. The presence of multiple paralogous TFs within a nucleus means that their binding must be understood, not simply as independent events, but as a function of competitive interactions that ultimately shape their regulatory landscapes. When TF paralogs are co-expressed in the same cells, they can compete with each other in two different ways to regulate gene expression. First, they can compete with each other for interacting with common protein partners. A well-known example comes from the bHLH family, where Myc and Mad1 compete with each other for interacting with Max, which in turn can also homodimerize [39]. The relative concentrations of these proteins, along with their affinities for each other, determine the relative ratios of Myc:Max, Max:Max, and Mad1:Max dimers in the cell, creating a highly tunable regulatory landscape. Second, when TF paralogs with shared intrinsic DNA preferences are co-expressed in the same cell, they can also compete for binding to the same targets in the genome. The Myc:Max, Max:Max, and Mad1:Max dimers all recognize the ‘CACGTG’ E-box motif but have opposing regulatory functions – Myc is a potent transcriptional activator, whereas Mad1 is a repressor [39]. Because they share the same core binding sequence, Myc and Mad1 compete for binding at ‘CACGTG’ E-box sites. In addition to the relative concentrations of these paralogs, competitive binding between them may also be modulated by subtle sequence-specific DNA preferences that differ between Myc:Max, Max:Max, and Mad1:Max [37]. The underlying DNA sequence therefore provides an additional regulatory layer, where evolutionary fine-tuning of binding sites can bias transcriptional outcomes toward activation (by Myc:Max) or repression (by Mad1:Max) by preferentially recruiting specific paralogs.

The outcome of DNA-binding competition between paralogous TFs depends on the relative nuclear abundance of each TF. If two paralogs truly have identical DNA binding activities, then whichever TF is more highly expressed will occupy more of the shared sites proportional to their concentrations. In reality, though, most paralogs are not identical to each other and often diverge in their preferences at certain nucleotide positions, DNA shape features and lower-affinity binding sites. These intrinsic specificity differences mean that each paralog may have slightly altered affinity at some sites versus others. As a result, when both TFs are co-expressed, the competition is sequence-specific, which means that a paralog may outcompete its paralog at lower concentrations. One of the most well-characterized examples of this comes from the yeast bHLH TFs Cbf1 and Pho4, which compete for binding to ‘CACGTG’ E-box motifs [40]. In phosphate-replete conditions, Cbf1 dominates most ‘CACGTG’ sites, while Pho4 remains cytoplasmic. However, upon phosphate starvation Pho4 is translocated to the nucleus, where it competes with Cbf1 for occupancy at phosphate-responsive genes [40]. The outcome of this competition directly determines which genes are activated by Pho4 and which remain occupied by Cbf1, highlighting the regulatory significance of TF competition. High-throughput in vitro competition genomic context PBM experiments have demonstrated that when the concentration of Pho4 is artificially increased relative to Cbf1, it can successfully displace Cbf1 at many shared binding sites [38]. However, at sites where Cbf1 displays a slight preference for specific flanking nucleotides, Pho4 fails to outcompete Cbf1 even at high Pho4 concentrations [38]. A comprehensive study of whole genome duplicated yeast paralogs has shown that deletion of several TFs indeed alter binding of their paralogs at individual gene promoters [12]. However, competition for DNA binding between TF paralogs has also been shown to constrain the evolution of an essential TF following gene duplication in unicellular yeast [41]. Consequently, it is unclear how widespread this phenomenon is in extant species. Given that the number of TF paralogs often increase with organismal complexity, it is therefore critical to understand how competition between TF paralogs constrained evolution of gene function following duplication in the past and how it adds to regulatory complexity in extant multicellular organisms.

Competition for DNA binding sites has been shown to be critically important even between unrelated TFs, such as the ubiquitous activator NF-Y and stage-specific repressor BCL11A [42]. Steric hindrance between them for closely spaced binding sites controls the developmental switch from fetal to adult hemoglobin expression [42]. However, competition is particularly well-characterized, and perhaps more widespread, among paralogous TFs that share conserved DBDs and recognize similar motifs. This is exemplified by the SP/KLF family, members of which are characterized by three C2H2-type ZFs at their C-terminus, and play crucial roles in embryonic development, cell differentiation and tissue-specific gene expression. Members of this family individually function as transcriptional activators (SP1, KLF1, KLF2, KLF5), repressors (KLF8, KLF10, KLF12, KLF15, KLF16) or both (KLF3, KLF4, KLF11), depending on the presence of specific activation and/or repression domains at the N-terminus [43]. However, multiple members of the SP/KLF family are often co-expressed in cells and compete for shared CACCC-box motifs. Compelling mechanistic evidence comes from studies of the KLF1-KLF3-KLF8 regulatory network in erythroid cells, where KLF3 directly represses KLF8 gene expression while KLF1 activates it, with both factors competing for binding to CACCC-box motifs in the KLF8 promoter [44,45]. This competitive binding creates an incoherent feed-forward loop where KLF1 directly activates KLF3 expression, but KLF3 then competes with KLF1 for shared target sites, establishing a self-limiting regulatory circuit that fine-tunes transcriptional output and prevents excessive KLF1-driven proliferation during stress [45]. Remarkably, competition between activators and repressors for DNA binding seems to be a common mechanism for precise transcriptional control by the SP/KLF family because multiple members (KLF4, KLF13, SP3 and SP4) have been shown to repress target gene expression by the ubiquitous activator SP1 in distinct cells and tissues [46–49]. While competitive binding dynamics have been characterized extensively within select families like SP/KLF, systematic investigation across the broader landscape of paralogous TF families remains largely unexplored, representing a critical gap in our understanding of transcriptional regulation.

Current computational frameworks for interpreting disease-associated noncoding variants operate under the implicit assumption that TF binding can be modeled as independent events, analyzing individual TF-DNA interactions in isolation without considering the competitive dynamics that govern regulatory element occupancy in vivo. However, the extensive characterization of competitive binding within the SP/KLF family, where activators and repressors routinely compete for shared CACCC-box motifs to establish precise transcriptional control, demonstrates that such competitive interactions are not merely secondary effects but fundamental determinants of regulatory outcomes. Given the pervasive co-expression of paralogous TFs across developmental stages and cell types, these competitive binding dynamics likely represent a widespread regulatory mechanism that remains systematically unexplored across the majority of paralogous TF families. A failure to account for these interactions may fundamentally limit our ability to predict the functional consequences of regulatory sequence variation and understand the molecular basis of transcriptional dysregulation in human disease.

TF paralogs have divergent preferences for lower-affinity DNA binding sites

The competitive binding dynamics described above raise a fundamental question: if paralogs share the same core motifs, how do they achieve distinct regulatory outcomes? An integrative study of 11 TF paralogs found that despite sharing the same core motif, each TF bound a unique subset of sites in vitro [37]. This apparently paradoxical observation has been resolved by recent technical advances that have significantly expanded our understanding of TF-DNA binding interactions. Several studies in the past decade have observed that nucleotides flanking motif matches in the genome alter TF occupancy [50–58]. Some of them have attributed this to particular DNA shape preferences of TFs [50,53]. However, it is not clear how DNA shape, which is ultimately determined by the underlying DNA sequence [59], correlates with TF binding affinities. On the other hand, traditional methods for characterizing TF-DNA interactions have focused primarily on identifying high-affinity binding sites, lacking the sensitivity to detect lower-affinity interactions that may distinguish paralog-specific binding. We recently developed PADIT-seq, using which we comprehensively measured TF binding preferences across all possible 10-bp sequences with unprecedented sensitivity [60]. By detecting hundreds of novel, lower-affinity binding sites, PADIT-seq revealed that motif models are inherently biased toward capturing high-affinity TF-DNA interactions while missing the far more numerous lower-affinity interactions. The expanded repertoire of lower-affinity binding sites revealed that high-affinity binding sites do not exist in isolation but are flanked by nucleotides that create additional, overlapping, lower-affinity binding sites [60]. These lower-affinity binding sites, though individually weaker, collectively play a crucial role in determining overall TF occupancy. The total binding affinity of a TF for a given genomic sequence, therefore, is not determined solely by its strongest site, but rather by the cumulative effect of all overlapping binding sites within the region.

The overlapping binding sites model provides a mechanistic explanation for the divergent genomic preferences observed among paralogous TFs. While paralogs share recognition of core high-affinity motifs, they often exhibit substantial differences in their preferences for lower-affinity binding sites. Differential preferences for lower-affinity sites create distinct patterns of overlapping binding sites flanking shared high-affinity sites. For example, when the paralogous TFs Pho4 and Cbf1 encounter the same high-affinity ‘CACGTG’ E-box motif in different genomic contexts, the flanking sequences create paralog-specific patterns of overlapping lower-affinity sites that ultimately determine which paralog has higher affinity for the DNA sequence (Figure 2) [60]. By incorporating the contribution of lower-affinity binding sites, the overlapping binding sites model provides a more accurate representation of TF occupancy in the genome and offers new insights into how subtle sequence variations between paralogous DBDs can fine-tune DNA binding specificity.

Figure 2:

Under low phosphate conditions, Pho4 enters the nucleus to compete with Cbf1 for genomic occupancy at CACGTG E-box elements. Pho4 outcompetes Cbf1 when nucleotides flanking CACGTG create more Pho4-specific overlapping binding sites.

Given that TFs recognize a far more extensive repertoire of lower-affinity binding sites than previously appreciated, an important question arises: how do TFs achieve sequence-specific recognition of such a vast collection of DNA sequences? Studies to date have largely focused on high-affinity TF-DNA interactions, but detailed structural analyses of lower-affinity TF-DNA complexes will be crucial to elucidate the molecular basis of these interactions. Beyond sequence-based preferences for lower-affinity sites, cooperative interactions with partner TFs represent an additional mechanism that modulates competitive binding dynamics between paralogs. Evidence from the Pho4-Cbf1 system demonstrates that cooperative binding with Pho2 can also influence which paralog ultimately occupies ‘CACGTG’ E-box motifs in the genome [40]. Systematic investigation of how cooperative TF-TF interactions modulate competitive binding dynamics represents an important frontier for understanding paralog-specific regulatory mechanisms.

Future Directions

As natural experiments in protein evolution, paralogous TFs have served as uniquely powerful experimental systems for dissecting the mechanistic basis of DNA binding specificity. Simultaneous measurement of binding affinities for hundreds of TF variants across multiple DNA sequences have revealed that members of the same DBD family can share identical high-affinity binding sites but differ greatly in selectivity, which is the ratio of affinity for preferred versus non-preferred sites [61,62]. Remarkably, mutations positioned distally from the DNA interface can allosterically modulate selectivity, not through direct alteration of DNA contacts but by shifting the TF’s folding and binding dynamics [62,63], representing an underappreciated source of specificity in TF-DNA interactions. Systematic studies across multiple TF families, including homeodomains [34], bHLHs [62] and forkheads [64], have demonstrated that mutations in non-DNA-contacting residues can significantly alter DNA binding specificity and selectivity. Similar insights have emerged from ancestral protein reconstruction studies of nuclear receptor paralogs [65], where three critical substitutions in the recognition helix were required to switch DNA binding specificities between the two DBDs [65]. However, they were functional only within a permissive evolutionary background [65]. These permissive mutations, positioned distally from the DNA contact interface [66], sufficiently elevated baseline binding affinity to enable subsequent recognition helix changes to redirect specificity without compromising functional activity [65].

The integration of high-throughput binding measurements with evolutionary reconstruction approaches provides a comprehensive framework for understanding how subtle sequence variations between paralogous TFs generate functional diversification. However, the molecular mechanisms by which these distal mutations influence DNA recognition specificity remain poorly understood. Future investigations must elucidate how amino acid changes distant from the DNA interface can allosterically modulate sequence-specific recognition. This will be crucial for predicting the functional impact of TF coding mutations in the human population.