Homeodomain complex formation and biomolecular condensates in Hox gene regulation

Abstract

Hox genes are a family of homeodomain transcription factors that regulate specialized morphological structures along the anterior-posterior axis of metazoans. Over the past few decades, researchers have focused on defining how Hox factors with similar in vitro DNA binding activities achieve sufficient target specificity to regulate distinct cell fates in vivo. In this review, we highlight how protein interactions with other transcription factors, many of which are also homeodomain proteins, result in the formation of transcription factor complexes with enhanced DNA binding specificity. These findings suggest that Hox-regulated enhancers utilize distinct combinations of homeodomain binding sites, many of which are low-affinity, to recruit specific Hox complexes. However, low-affinity sites can only yield reproducible responses with high transcription factor concentrations. To overcome this limitation, recent studies revealed how transcription factors, including Hox factors, use intrinsically disordered domains (IDRs) to form biomolecular condensates that increase protein concentrations. Moreover, Hox factors with altered IDRs have been associated with altered transcriptional activity and human disease states, demonstrating the importance of IDRs in mediating essential Hox output. Collectively, these studies highlight how Hox factors use their DNA binding domains, protein-protein interaction domains, and IDRs to form specific transcription factor complexes that yield accurate gene expression.

Graphical Abstract

1.0. Introduction

The highly conserved Hox gene family has long fascinated geneticists, developmental biologists, and evolutionary biologists [1–3]. Early studies in Drosophila melanogaster revealed a series of linked genetic mutations that resulted in dramatic homeotic transformations of appendages and segments along the anterior-posterior (A-P) axis [4, 5]. Subsequent molecular studies showed that each Drosophila Hox gene has a highly similar 180 nucleotide signature sequence called the homeobox that encodes a 60 amino acid homeodomain that binds DNA in a sequence-specific manner [6–8]. Hox genes are conserved across metazoans and often found clustered along the chromosome, albeit the number of Hox genes frequently differs between animal species due to either individual gene duplication or gene loss within a Hox cluster or the duplication/loss of an entire Hox gene cluster (Figure 1A) [9, 10]. Moreover, the expression of Hox genes along the A-P axis generally correlates with their order found within the cluster, and Hox genes have been categorized into distinct groups based on their role in anterior versus central versus posterior fate specification (Figure 1A) [11]. Advances in gene targeting and genomic editing technologies led to an explosion in studies focused on altering Hox gene function in numerous different animal species [10, 12, 13]. Cumulatively, studies using model organisms as well as other species have consistently demonstrated how changes in Hox gene number and expression patterns along the developing A-P axis can instruct the formation of distinct cell fates and specialized morphological structures.

Figure 1: The clustered Hox genes encode conserved homeodomain transcription factors with characteristic protein regions across phyla.

(A) Schematic showing the stereotypic chromosomal gene organization of Hox factors with the order within the cluster tied to their expression along the anterior-posterior axis. Drosophila contains a split chromosomal cluster on the same chromosome that separates the Hox genes into the Antennapedia cluster consisting of Lab, Pb, Dfd, Scr and Antp and the Bithorax cluster containing Ubx, Abd-A, and Abd-B. Vertebrates typically contain several Hox clusters that often lack a subset of the Hox genes due to gene loss. Here, we show a single vertebrate Hox cluster with a full complement of genes numbered from Hox1 to Hox13. Color coding denotes the approximate developmental relationship from ancestral Hox cluster, including duplication events of various genes, and the Hox genes can be broadly assigned into anterior, central, and posterior groups based on their expression and role in anterior-posterior patterning.

(B) The clustered Hox genes and many non-Hox homeodomain proteins bind highly similar AT-rich monomer sequences in vitro. Representative PWMs of monomer Hox motifs for Scr and Ubx and the non-Hox homeodomain binding motifs for Distal-less (Dll) and GSX2 were taken from SELEX experiments uploaded to CisBP (http://cisbp.ccbr.utoronto.ca).

(C) Schematics of a generic Hox protein and two Drosophila Hox factors with the homeodomain, short linear interaction motifs (SLiMs) and intrinsically disordered regions highlighted. Hox factors use various combinations of these sequences (i.e. Scr vs. Ubx) for specificity. IDRs in Scr and Ubx are annotated based on predicted regions on UniProt. Lab = Labial; Pb = Proboscopedia; Dfd = Deformed; Scr; Sex Combs Reduced; Antp = Antennapedia; Ubx = Ultrabithorax; Abd-A = Abdominal-A; Abd-B = Abdominal-B; Dll = Distalless; GSX2 = GS Homeobox 2; IDR = Intrinsically Disordered Region; SLiM = Short Linear Interaction Motif; HD = Homeodomain; YPWM = hexapeptide motif; UA = the conserved UbdA motif. Created with BioRender.com

Given their central role in both the development and evolution of distinct morphological structures, a great deal of effort has been spent trying to understand how individual Hox genes accurately control the gene regulatory networks required for specific cell fates. However, biochemical studies revealed that in sharp contrast to their ability to direct distinct in vivo fates, the Hox homeodomain proteins largely bind the same AT-rich DNA sequences in vitro (Figure 1B) [14–17]. This problem is further exacerbated by the fact that metazoans encode many additional non-Hox homeodomain transcription factors that bind highly similar AT-rich DNA sequences (Figure 1B). For example, the human genome encodes approximately 200 homeodomain transcription factors, of which only 39 are the classic Hox genes that specify fates along the A-P axis [18, 19]. Taken together, these findings present a paradox: How can a set of transcription factors with highly similar DNA binding activities achieve sufficient target gene specificity to direct the development of fundamentally different cell fates and morphological structures?

While many mechanisms are likely to contribute to Hox specificity, including post-translational modifications [20, 21], binding to coactivator and corepressor proteins [20, 22], and cellular chromatin status [23], this review focuses on the mechanisms that alter and enhance Hox DNA binding specificity. Over the past three decades, molecular studies have revealed that in addition to the conserved homeodomain that directly contacts DNA, Hox factors contain different combinations of short linear interaction motifs (SLiM) that bind other proteins [24–26] and intrinsically disordered regions (IDRs) that alter the biophysical properties of Hox proteins (Figure 1C). Among the Hox SLiMs identified to date, several have been shown to specifically interact with additional transcription factors, especially other homeodomain proteins, and biochemical and genomic studies have revealed how homeodomain transcription factor complexes increase Hox DNA binding specificity [27]. While less is known about the role of IDRs in Hox function, recent work on IDRs has demonstrated a role in localizing proteins to nuclear sub-compartments [28, 29], providing a new mechanism for how Hox proteins can be concentrated to increase binding to low-affinity binding sites within target gene cis-regulatory modules (CRMs). The distinct combinations of SLiMs and IDRs within Hox proteins provides key insight into how Hox proteins may gain sufficient regulatory specificity during development (Figure 1C). Thus, this review will focus on the molecular mechanisms that target Hox proteins to specific DNA sequences during development.

2.0. Hox complex formation with the Pbx and Meis homeodomain proteins.

While we still do not have a complete answer to how Hox transcription factors achieve accurate DNA target specificity, several key concepts have begun to emerge. Chief among these mechanisms is that Hox factors use SLiMs to form a variety of different transcription factor complexes on DNA with the Pbx (Extradenticle (Exd) in Drosophila) and Meis (Homothorax (Hth) in Drosophila) families of transcription factors [30]. Pbx and Meis are highly conserved members of the Three Amino Acid Loop (TALE) homeodomain family of proteins. Molecularly, the Pbx and Meis factors can interact with each other through highly conserved N-terminal domains, as well as with Hox factors to form larger transcription factor complexes on DNA. However, it is important to note, Pbx and Meis factors also have Hox-independent functions and can also interact with additional non-Hox transcription factors [31].

The role of the Pbx and Meis factors as Hox cofactors has been well described in the literature with several excellent reviews highlighting the many aspects of how Pbx and Meis impact Hox function [23, 27, 32, 33]. Here, we focus on four mechanisms that reveal how Pbx and Meis enhance Hox transcription factor DNA binding specificity. First, the Pbx TALE homeodomain directly interacts with a short hexapeptide motif (typically containing a YPWM core) found N-terminal to the homeodomain of Hox factors [30]. This interaction, which is thought to be relatively weak off DNA, results in the formation of cooperative Pbx/Hox complexes on DNA through adjacent Pbx/Hox binding sites (Figure 2A). Importantly, the added DNA sequence requirements needed to bind both Pbx and Hox factors as well as the direct protein-protein interactions enhance the DNA binding affinity and specificity relative to other homeodomain factors (Figure 2A). For example, while Hox factors bind similar monomer DNA sites as many other homeodomain factors, such as Gsx2 (Figure 1B), Gsx2 does not have a hexapeptide interaction motif, has not been shown to form complexes with Pbx, and does not enrich for Pbx/Hox binding sites in genomic assays [14–16, 34]. In contrast, Hox factors have either a well-defined or atypical hexapeptide motif that functions as a SLiM to bind Pbx and thereby form heterodimers on DNA [30] (Figure 2A). Highlighting the importance of the hexapeptide motif for Pbx/Hox genomic binding, a ChIP-seq study for the Sex combs reduced (Scr) Hox factor in Drosophila revealed that flies engineered with a Scr-YPWM mutation were homozygous lethal and only enriched for monomer Hox sites, whereas wild-type Scr strongly enriched for Exd/Hox dimer sequences [35]. Intriguingly, a subset of Hox factors contain more than one SLiM sequence that directly interacts with Pbx/Exd. For example, the Ultrabithorax (Ubx) and Abdominal-A (Abd-A) Hox factors contain both the generic hexapeptide sequence as well as a UbdA motif (UA motif, Figure 1C) that can also mediate interactions with Exd [36–38]. Moreover, a subset of vertebrate Hox factors similarly contain multiple Pbx interacting motifs [39, 40]. While current studies favor the idea that Hox factors with multiple Pbx/Exd interaction domains function redundantly to promote Hox heterodimer binding, many of these studies have relied upon ectopic expression assays and only assessed DNA binding to specific CRM sites. Hence, genomic studies using animals with specific Hox factor SLiM mutations in endogenous loci and unbiased SELEX-seq assays are needed to determine if the Hox factors use distinct Pbx/Exd interacting peptides to alter and/or diversify the sequence binding selectivity of the heterodimer complex on DNA.

Figure 2: Mechanisms of Hox factor DNA binding specificity in complex with the Pbx and Meis TALE homeodomain proteins.

(A) All Hox factors can interact with Pbx/Exd to increase DNA binding sequence specificity and affinity. In contrast, most non-Hox homeodomains do not similarly interact with Pbx on DNA.

(B) Hox factors largely bind similar monomer sites in vitro, but latent specificity differences are accentuated through interaction with Exd. Preferred Exd/Hox heterodimer sequences were defined by Slattery et al. [41]

(C) Posterior Hox factors interact with Meis/Hth proteins to promote differential sequence recognition between anterior and posterior Hox factors.

(D) The most common Hox multiplex organization uses nearby Meis binding site and adjacent Pbx/Hox heterodimer site sto increase binding specificity of Hox factors at target gene CRMs. Note, flexible configurations of these motifs are possible with distinct combinations of homeodomain binding sites potentially contributing to DNA binding specificity. Pbx1 = PBX Homeobox 1; Exd = Extradenticle; Meis1 = Meis Homeobox 1. Created with BioRender.com

Second, while all Hox factors can bind DNA as both monomers and with Pbx/Exd, heterodimer binding was found to accentuate differences in DNA binding site preferences between Hox factors (Figure 2B). This concept, which is called latent specificity, was revealed when Slattery et al performed a comprehensive unbiased DNA binding site selection assay (SELEX-seq) for eight Drosophila Hox factors as monomers and in combination with Exd [41]. Importantly, comparative analysis of the bound monomer versus Exd/Hox dimer sequences revealed that the differences in binding preferences between Hox factors increased when in complex with Exd relative to monomer binding alone (Figure 2B). To better understand how Exd/Hox interactions invoke latent specificity, computational analysis based on DNA shape revealed that when in complex with Exd, the anterior Hox factors prefer DNA sites with narrower minor groove widths relative to the central/posterior Hox factors [41, 42]. This computational prediction was supported by prior structural, biochemical, and genetic studies showing that Scr utilizes conserved amino acid residues in the N-terminal arm of its homeodomain to make specific contacts to DNA sequences with a narrow minor groove [43, 44]. To formally test this idea, SELEX-seq and reporter assays using Scr proteins containing specific mutations in the N-terminal arm residues that contact the narrow minor groove demonstrated how paralog-specific residues are required for Exd/Scr binding to sequences with this DNA shape feature [42]. Moreover, selectively inserting these Scr residues into the N-terminal arm sequences of Antennapedia (Antp), a more posterior Hox factor that does not efficiently bind DNA with a narrow minor groove, was shown to be sufficient to confer this DNA binding site preference in vitro and target gene expression activity in vivo [42]. Thus, these studies demonstrate how Pbx/Hox heterodimer complexes can further differentiate Hox DNA binding activities relative to Hox monomer binding.

Third, the Meis/Hth TALE factors form complexes with a subset of Hox factors on DNA (Figure 2C). While the molecular mechanisms and the Hox sequences underlying the interactions between Meis and Hox factors are not well understood, DNA binding assays revealed that Meis cooperatively binds adjacent Meis/Hox sites with vertebrate members of the posterior Hox9-13 class of factors [45]. Consistent with these findings, ChIP-seq studies for MEIS1 and HOXB13 in prostate cells revealed enrichment of adjacent MEIS and HOX sites at many genomic loci [46]. In contrast, the Hoxb4, Hoxb6, Hoxa7, and Hoxb8 anterior/central Hox factors failed to similarly bind such Meis/Hox sites in vitro, suggesting Meis factors selectively bind heterodimer DNA sites with only posterior Hox factors (Figure 2C)[45]. In Drosophila, Hth was also shown to complex with a subset of Hox factors on adjacent binding sites to mediate both gene repression and activation. In particular, Hth cooperatively binds adjacent Hth/Hox DNA sites with Ubx and Abd-A to repress Distal-less (Dll) expression and leg development in abdominal segments [47, 48]. In addition, Hth/Abd-A cooperatively bind and activate a CRM that ultimately promotes EGF signaling via similar adjacent Hth/Hox sites [49, 50]. Interestingly, Ubx and Abd-A are members of the central Hox class (Figure 1A), and the vertebrate central Hox factors (Hoxb6, Hoxa7, and Hoxb8) do not cooperatively bind with vertebrate Meis in DNA binding assays [45]. In addition, Antp, which is also a member of the Drosophila central Hox factors, failed to regulate these targets through Hth/Hox binding sites [49–51]. Hence, future studies are needed to determine which Hox factors are capable of cooperatively binding adjacent sites with Meis/Hth and to define the paralog-specific mechanisms underlying the formation of these complexes.

Fourth, Pbx, Meis, and Hox factors can form larger transcription factor complexes on DNA (Figure 2D). Since Pbx and Meis interact via N-terminal domains separate from their C-terminal homeodomains [52], the Pbx and Meis binding sites can be found at variable distances and orientations relative to each other [27]. In complex with Hox factors, the most common trimeric protein complex contains adjacent Pbx/Hox sites with a more distant Meis binding site [20]. Such complexes and motif enrichment profiles have been observed in both vertebrate and Drosophila studies [53–56]. Moreover, a SELEX-seq assay using Drosophila proteins demonstrated that each Hth binding site orientation was permissible relative to the Exd/Hox sites, although one orientation preferred a shorter spacer distance between the Hth and Exd/Hox sites than the other orientation [57]. Additional studies in Drosophila revealed that even larger Hox complexes can be observed as Hth/Hox and Exd/Hox sites separated by 7bps can mediate the formation of functional tetrameric complexes [47, 51]. Thus, these studies highlight how distinct combinations of Hox, Pbx/Exd, and Meis/Hth sites can be utilized to recruit larger homeodomain transcription factor complexes that mediate specific transcriptional outcomes.

2.1. Hox complex formation with other homeodomain transcription factors

While genetic, biochemical, and genomic studies strongly support the idea that the Pbx and Meis factors are widespread Hox co-factor proteins, studies have begun to uncover that additional transcription factors contribute to Hox DNA binding specificity. For example, the Engrailed (En) homeodomain protein was shown to cooperatively bind DNA with the Ubx and Abd-A Hox factors but not the Antp Hox factor [47]. Intriguingly, subsequent molecular studies revealed that Abd-A requires its UbdA motif to mediate interactions with En, suggesting that this motif may participate in binding both the Pbx and En homeodomain proteins [36]. In addition to En, recent ChIP-seq and biochemical pull-down studies have shown that the Distal-less (Dll) homeodomain factor may selectively function as a Hox co-factor protein with Scr during leg development. In particular, Feng at al found that Scr and Dll bind to many of the same genomic sequences containing two TAATTA sequences spaced 3bp apart in the leg imaginal disc in the first thoracic segment (T1) [35]. Intriguingly, while Scr is expressed throughout the leg imaginal disc, the Exd/Hth factors and the Dll factors are expressed in largely mutually exclusive domains along the proximal-distal axis. Hence, these studies suggest that Scr forms distinct transcription factor complexes with different homeodomain factors to regulate distinct target genes during T1 leg development [35]. Lastly, additional bimolecular fluorescent complementation (BiFC) assays and mass-spectrometry studies have revealed that Hox factors can interact with many, perhaps even hundreds, of different transcription factors, including many homeodomain proteins [26, 58, 59]. While the mechanisms and impact of these potential interactions on DNA binding are largely unknown, these studies suggest that the formation of larger Hox transcription factor complexes with different partner proteins may be a critical, widespread mechanism underlying how Hox factors achieve target gene specificity.

2.2. The role of low-affinity binding sites in Hox specificity

The above studies support a model that Hox factors use SLiMs to form complexes with other homeodomain transcription factors to increase DNA binding specificity, suggesting that CRMs use different combinations of AT-rich DNA sequences to recruit specific Hox complexes. Intriguingly, studies on several known Hox-regulated CRMs have revealed an additional concept: Binding sites that yield Hox-specific output often contain clusters of sub-optimal or low-affinity sites (Figure 3A). In fact, recent evidence suggests that low-affinity sites are generally important for proper CRM function, as they promote target specificity between related transcription factors, and increasing the binding affinity of such sites can result in CRM mis-regulation and developmental defects [60–67]. One of the best examples of these principles are the shaven-baby (svb) CRMs that mediate Ubx-specific activation using multiple low-affinity Exd/Hox sites to regulate gene expression required for trichome patterns in Drosophila [62, 68]. Importantly, changing these sequences from low- to high-affinity resulted in the mis-regulation of svb CRM activity into additional segments in a Hox-dependent manner, supporting the idea that high-affinity sites are likely to be bound and regulated by many Hox factors, and thus are less specific to individual Hox factors (Figure 3B) [62, 69]. Consistent with this idea, a general computational approach called No Read Left Behind (NRLB) demonstrated a general inverse relationship between Hox/Exd binding affinity and Hox/Exd binding specificity [69]. Moreover, several other CRM studies similarly reveal how low-affinity sites can result in cell- and or segment-specific output, whereas high-affinity sites result in altered and/or expanded patterns of gene expression [50, 70–72].

Figure 3: The role of low-affinity sites and Hox transcriptional specificity.

(A) Sequence differences between low- and high-affinity Exd/Hox sites promote the binding of specific versus pan Hox factors. Note, this example highlights a Ubx-specific low affinity Exd/Hox regulatory element whereas high affinity sites were found to be regulated by many Hox factors. Sequences were defined and tested by Rastogi et al. [69]

(B) Presence of low-affinity Exd/Hox sites in vivo directs select expression of the svb target gene by Ubx. Converting the site into a high-affinity motif resulted in the mis-regulation of svb CRM activity due to more generalized Hox activation [62, 69].

(C) The wild-type Ubx proteins was found to be highly concentrated at transcriptional hubs in Drosophila nuclei whereas a DNA binding deficient Ubx protein was diffusely localized throughout the nucleus. svb = shavenbaby [76, 77]. Created with BioRender.com

Relying upon low-affinity binding sites to mediate cell-specific gene regulatory outcomes that are essential for proper development comes with a caveat: DNA-transcription factor binding events are often transient with DNA binding and dissociation occurring over a relatively short time-period [73–75]. Hence, high transcription factor concentrations are required to reproducibly bind and regulate CRMs using low-affinity sites. However, transcription factors are frequently expressed at relatively low cellular levels, raising the question of how low-affinity CRM binding sites can reproducibly recruit the required transcription factors and cofactors to function efficiently? High-resolution imaging methods have begun to answer this question. For example, further investigation of the svb CRMs showed that Ubx is not uniformly distributed across the nucleus (Figure 3C) [76]. Rather, Ubx is locally concentrated in specific nuclear subregions, and these regions of high Ubx concentration correspond to regions where the Hox cofactor Hth is concentrated. Tsai et al further found that Ubx and Hth concentrations are high at the sites of active transcription of the svb locus [76]. In fact, svb CRMs on homologous chromosomes as well as transgenic svb CRMs reporters inserted on different chromosomes colocalized with one another at sites of high Ubx concentration [76]. High Ubx concentrations located in close proximity to the svb CRMs with clustered low-affinity sites may provide an avidity effect, whereby Ubx molecules that dissociate from one site would likely have an increased probability to bind other nearby low-affinity sites. Thus, CRMs with clustered sites are likely to play a key role in the formation of “transcriptional hubs” that yield cell-specific outputs.

Several lines of evidence suggest that the formation of these active transcriptional hubs on the svb CRMs are directly dependent on Ubx interactions with binding sites. First, analysis of flies expressing a DNA binding deficient Ubx protein shows a uniform nuclear distribution of the transcription factor in contrast to the regions of high Ubx and Hth concentration seen with the wild-type factor (Figure 3C) [76]. Furthermore, when a redundant Ubx-regulated CRM from the svb locus was deleted, Ubx concentrations around the actively transcribed region were significantly decreased [77]. This resulted in overall decreased transcription from the svb locus, and increased sensitivity to stress with phenotypic defects appearing at increased temperatures [77]. Finally, insertion of another copy of the svb CRM onto a separate chromosome not only rescued the phenotypic effects of the deletion, but also resulted in increased local Ubx concentrations and transcription at the svb locus [77]. These data indicate that CRMs and gene loci on separate chromosomes can function cooperatively to promote formation of nuclear regions with increased transcription factor and cofactor concentration. These findings raise a new biological question: What are the mechanisms underlying how specific transcription factors are concentrated within specific nuclear sub-compartments?

3.0. Biomolecular condensates, IDRs, and transcriptional regulation

Increasingly, studies have revealed that the nucleus is highly heterogeneous, with transcription factors, such as the Hox factors, often found in discrete, concentrated nuclear sub-compartments (Figure 4). Intriguingly, biophysical and cellular studies have led to the identification, characterization, and functional analysis of membraneless nuclear sub-compartments with particular emphasis on the notion that liquid-liquid phase separation can concentrate proteins including transcription factors and the transcriptional machinery and thereby define transcriptionally active chromatin [78–83]. The observation of these subdomains is intriguing because it suggests a rather intuitive model for transcriptional regulation, where nuclear sub-compartments enriched for transcription factors, the mediator complex, and RNA polymerase II can interact with specific chromatin regions (promoters and distal enhancers).

Figure 4: The role of nuclear sub-compartments in transcriptional regulation.

Schematic at left shows a graphic representation of a theoretical “homogenous” nucleus with proteins diffusely found throughout the nucleus. Schematic in middle shows a representation of a heterogenous nucleus based on high resolution imaging. The heterogenous nucleus contains distinct regions of concentrated proteins, often called biomolecular condensates, that typically include specific transcription factors and the transcriptional machinery (i.e. RNA polII, the mediator complex, etc). Close-up view of these nuclear condensates, which rely upon liquid-liquid phase separation of proteins due to weak protein-protein intermolecular interactions via IDRs, reveals a Hox multimeric condensate with Pbx/Exd and Meis/Hth and a Hox monomeric condensate. In this model, the high local protein concentrations increases the probability of Hox binding to CRMs, and thereby the activation of specific target genes within each nuclear sub-compartment. RNA pol II = RNA polymerase II; Hth = Homothorax. Created with BioRender.com

Though we focus on the implications of nuclear condensates on the regulation of transcription, and particularly on how they impact the function of Hox transcription factors, it should be noted that the identification of membraneless compartments that facilitate compartmentalization and concentration of components required for specific cellular functions is not new. The role of the nucleolus in ribosomal formation is a prominent example. In fact, these compartments have been implicated in a variety of nuclear processes including heterochromatin formation, DNA replication, DNA repair, transcriptional regulation, and RNA processing [83–87]. Before moving forward, it must also be addressed that just as there are many different types of sub-compartments found throughout the cell, there are multiple different and sometime conflicting terms used to describe them. Recent studies, especially those focused on transcriptionally active sub-compartments, have used the term biomolecular condensates; however, other terminology includes chromatin sub-compartments, membraneless organelles, nuclear bodies, transcriptional factories, and transcriptional hubs [81, 83]. Some debate exists on the precise definition of these terms, for example some use the term condensate to generally refer to all such sub-compartments, while other groups hold that condensates imply the presence of liquid-liquid phase separation. For the purposes of this review, we will use the terminology defined by Sabari et al: Namely, biomolecular condensates refer to a membraneless cellular compartment where specific biomolecules are concentrated and which are composed of higher-order assemblies of biomolecules held together by multiple, dynamic, weak interactions [83]. An overall key feature of these condensates is their capacity to alter biomolecule concentrations, such that biomolecules required for a particular process facilitate the formation of large multicomponent complexes by inclusion of necessary components and exclusion of others.

A general model has emerged from these studies in which a transcription factor’s location within the nucleus is dictated by two types of domains; a highly structured DNA binding domain (DBD) and an often unstructured regulatory domain made of repetitive amino acid sequences referred to as intrinsically disordered regions (IDRs) [88, 89]. In this model, the IDR functions to promote condensate formation, whereas the DBD directs condensate localization to specific genomic regions. Put another way, the sequence specificity of the DBD results in transcription factor binding to CRMs within the genome, whereas the IDRs from large numbers of transcription factor molecules and the transcriptional machinery weakly interact to form biomolecular condensates at specific genomic regions (Figure 4) [88, 90–92]. Lastly, histone modifications and the factors that “read” these marks also play key roles in condensate formation and specificity. For example, BRD4 interacts with acetylated nucleosomes and promotes formation of active condensates, while the factor HP1α binds nucleosomes with heterochromatin marks and drives formation of heterochromatin condensates [82, 93].

The ability of transcription factors to form condensates via their IDRs has implications as a possible mechanism for achieving target specificity. As has been previously described, even high confidence information about a transcriptions factor’s preferred DNA binding site is usually insufficient to predict in vivo targets. Further, related factors with nearly identical in vitro binding specificities often bind different targets when expressed in the same cell types. The in vivo and in vitro binding preferences of two yeast transcription factors that contain IDRs were analyzed, revealing the surprising contribution of the IDR to in vivo target selection [94]. The zinc finger transcription factor, Msn2, binds sites containing the 5’-AGGGG-3’ motif, which was confirmed in chromatin binding profiles. However, only a subset of promoters encoding that motif were bound by Msn2, and related transcription factors that bind the same core motif interact with a different subset of promoters. The importance of the DBD and the IDR of Msn2 were then tested by assaying the DNA binding profiles of a DBD-only and a DBD deficient protein. While the DBD-only construct bound regions enriched for the core AGGGG motif, the binding profile differed significantly from that of full-length Msn2. On the other hand, the DBD deficient construct localized to most of the same promoters as the full-length transcription factor, although enrichment of the AGGGG motif was no longer observed. Similar findings were also observed with the yeast transcription factor, Yap1 [94], highlighting the potential role of IDRs in targeting transcription factor proteins to specific nuclear sub-compartments.

3.1. Hox transcription factors, IDRs, and condensate formation

Many investigations of transcriptionally active biomolecular condensates focused on the key role played by IDRs in factors like MED1 and BRD4 in driving the formation of liquid-liquid phase separations and promoting transcriptional activation [82, 88]. Transcription factor activation domains consisting of IDRs can take several different forms including acidic, Proline, Glutamine, or Serine/Threonine-rich regions [89, 95, 96]. The importance of IDR-containing activation domains for the function of Hox factors has been well established. For example, early functional domain mapping assays of the HOXD4 transcription factor revealed that in addition to its DNA binding domain, a proline-rich domain with characteristics of an IDR was necessary for transcriptional activation [97]. Interestingly, the Ubx Hox factor also contains IDRs that contribute to in vitro DNA binding affinity and its regulatory function [98, 99]. However, to our knowledge, the ability of the Ubx IDRs to form biomolecular condensates and their role in regulating known Ubx target genes, such as svb, has yet to be tested.

Direct evidence that IDRs are not only essential for Hox gene function, but also contribute to biomolecular condensates and liquid-liquid phase separation comes from investigations into a set of inherited diseases in humans. These diseases comprise a group of more than 20 disorders that result from the expansion of short, repetitive DNA sequences encoding amino acid repeats, typically of alanine or glutamine [100–102]. 15 of these disease-associated repeats were found to be in nuclear proteins, most being transcription factors [100]. Of particular note, expansion of an alanine repeat in HOXD13 has been implicated in the hereditary limb malformation disorder, synpolydactyly [103]. More recent work using a synpolydactyly mouse model has shown that the polyalanine region in HOXD13 is part of an IDR, which can drive liquid-liquid phase separation, and the disease-associated expansion of alanine repeats enhanced phase separation and altered the composition of HOXD13 condensates [28]. More specifically, compared to wild-type HOXD13 condensates, those created by the HOXD13 variant with the expanded alanine repeats, less effectively recruited the mediator complex and RNA pol II. Thus, the disease-associated allele is less effective at activating transcription [28]. Consistent with these findings both wild-type and disease-associated HOXD13 proteins have very similar DNA binding profiles as defined by ChIP-seq assays, but their transcriptional profiles are clearly different [28]. Importantly, repeat expansions in other transcription factors implicated in disease, such as HOXA13, RUNX2, and TBP, were also shown to alter both phase separation and condensate composition [28]. Taken together, these findings suggest that pathological TFs with expanded IDRs share two key features: an enhanced capacity to undergo liquid-liquid phase separation and decreased target gene transcription.

Further evidence that altering Hox condensate formation contributes to human disease comes from investigation of cancer/tumorigenesis. It has been noted that many cancers have chromosomal translocations between genes that encode IDRs that are either associated with transcription factors or chromatin binding proteins. For example, a translocation that fuses the phenylalanine/glycine-rich IDR of a nucleoporin gene with one of several transcription factors has been implicated in leukemias [104–107]. Of note, a translocation that creates a fusion between the IDR from NUP98 with HOXA9 has been extensively studied in the context of leukemias [105, 107]. The NUP98-HOXA9 fusion is not only required for leukemic transformation but also can drive the formation of liquid-liquid phase separations [29]. In fact, this NUP98-HOXA9 protein has increased chromatin occupancy compared to HOXA9 alone and produces a binding profile with broad characteristics suggestive of “super-enhancer” formation. Analysis of chromatin looping using Hi-C showed that the NUP98-HOXA9 fusion showed enrichment for key proto-oncogenes. Further experiments showed that a fusion protein created by replacing the nucleoporin IDR with an unrelated IDR could also drive liquid-liquid phase separated condensates and produce similar increases in DNA binding and target gene activation [29].

Despite a relatively small number of studies investigating biomolecular condensates and Hox transcription factors, there is clear emerging evidence that at least some Hox transcription factors can contribute to such condensates by promoting liquid-liquid phase separations via weak interactions between IDRs. Furthermore, the ability of Hox factors to participate in condensates has obvious implications for pathogenesis of human diseases.

4.0. Future Directions

New evidence about the mechanisms by which Hox factors achieve specificity, regulate target genes, and both benefit from and participate in the formation of biomolecular condensates provides exciting new avenues for study, while also suggesting potential elegant solutions to long standing questions. It will be interesting to investigate whether regions of high Ubx concentration observed in Drosophila truly represent biomolecular condensates, whether they are mediated by IDRs within the Ubx protein or its cofactors, and to what extent these IDRs contribute to Hox target specificity. Further, it is interesting to consider that many Hox factors can function as both transcriptional activators and repressors. While this review focused mainly on the incorporation of Hox factors into condensates associated with activation and active transcription, it is worth noting that evidence of repressive condensates has also been described [108, 109]. Investigation of Hox factors in active versus repressive condensate formation could provide further insight into how the factors achieve regulatory specificity on different targets.

Highlights.

• Hox complex formation with other homeodomain factors increase DNA specificity

• Different Hox complexes have distinct DNA binding site requirements

• Low-affinity DNA sites increase Hox binding specificity

• Hox factors are concentrated within distinct nuclear sub-compartments

• Hox factor intrinsically disordered regions (IDRs) form biomolecular condensates

• Hox factors with altered IDRs have been associated with human diseases