Decoding the mechanisms of cooperative DNA binding by the Paired-like homeodomain family

Brittany Cain; Connor Wasmund; Fiona C Rowan; Brian Gebelein

doi:10.1038/s41467-026-76294-w

. 2026 Aug 3;17:9315. doi: 10.1038/s41467-026-76294-w

Decoding the mechanisms of cooperative DNA binding by the Paired-like homeodomain family

Brittany Cain ^1,^✉, Connor Wasmund ^1,⁴, Fiona C Rowan ², Brian Gebelein ^1,^3,^✉

PMCID: PMC13530268 PMID: 42675056

Abstract

The 36 Paired-like homeodomain transcription factors in humans are required for the development of many cell types, tissues, and organs as missense variants in 24 Paired-like genes have been associated with numerous diseases and developmental disorders. How these factors identify distinct genomic targets using highly similar DNA binding domains is not fully understood. Here, we focus on determining how the human Paired-like homeodomain factors gain DNA binding specificity by cooperatively binding palindromic sites spaced three base pairs apart (P3 site). Through structural, biochemical, and bioinformatic approaches, we define 11 rules that describe homeodomain residues that are critical, permissive, and inhibitory to cooperativity on the P3 site. Applying these rules, we successfully alter the cooperative behavior of Paired-like factors, identify residues that prevent the related Antennapedia class of homeodomains from binding cooperatively, and predict that thirty-eight disease-associated missense variants across ten Paired-like proteins alter cooperativity. Using quantitative DNA binding assays, we confirm eleven of twelve of these disease-associated variants impact cooperativity but not DNA binding affinity. These findings reveal the importance of cooperativity in defining DNA binding specificity and highlight how missense variants associated with as many as fifteen different diseases can selectively disrupt cooperative DNA binding.

Subject terms: DNA-binding proteins, Transcription factors

The authors show that the cooperative DNA binding behavior of 36 Paired-like factors is driven by 11 homeodomain residue positions. These mechanistic insights explain the functional consequences of 38 pathogenic missense variants across the Paired-like family.

Introduction

The homeodomain (HD) family is one of the largest families of transcription factors (TFs) with over 200 members in the human genome¹. HD proteins are classified by their 60 amino acid helix-turn-helix domain, most of which bind highly similar AT-rich DNA sites^1–4. However, HD TFs typically bind distinct genomic targets and regulate different developmental pathways. Moreover, HD TFs have the most disease associated alleles reported in the Human Gene Mutation Database (HGMD) among all TF families⁵, with numerous HD missense variants associated with a variety of diseases and birth defects. The way in which HDs acquire sufficient DNA binding specificity to regulate distinct biological outcomes, and how many of the pathogenic missense variants impact HD function is not well understood. Although computational tools have been developed to predict the consequences of single missense variants, a comprehensive analysis of 42 such prediction tools found that they largely fail to accurately identify variants that altered HD specificity⁵, emphasizing the importance of elucidating the specificity-gaining mechanisms.

HD TFs can be categorized into distinct subfamilies based on sequence homology and the presence of additional conserved domains and motifs⁶. For example, the Paired/PAX subfamily of HD TFs are defined by having a second DNA binding domain, known as the Paired (Prd) domain, and can thereby bind three types of DNA motifs using either the Prd domain, the HD, or both domains⁷. In contrast, the Paired-like factors, which constitute a family of 36 human genes that closely share a HD with the PAX factors but lack an associated Paired domain, encode only a single DNA binding domain. Thus, the Paired-like factors, much like the even larger Antennapedia (ANTP) subclass of HD factors that similarly lack a second DNA binding domain, must increase specificity by other mechanisms⁶.

One mechanism employed by TFs to increase their DNA binding specificity is the formation of complexes that bind longer recognition sequences. A well characterized example is the HOX factors within the ANTP subclass that specify distinct anterior-posterior fates along the body plan. HOX factors have relatively low DNA sequence specificity when bound alone but have distinct, increased specificity when complexed with PBX factors⁸. Previous work also revealed that some members of the Paired-like HDs form homo and heterodimers on a TAAT – NNN – ATTA P3 site^2,9–13. Moreover, genomic binding studies for the PHOX2A¹⁴, PHOX2B^15,16, ALX4^17,18, and CRX¹⁹ factors provided direct evidence that each of these Paired-like HD factors bind P3 sites in vivo. To better understand which Paired-like HDs cooperatively bind DNA, we previously developed the Cooperativity Predictor pipeline to analyze HT-SELEX data to provide evidence that at least 8 of 23 Paired-like factors bind cooperatively to the P3 site⁹. Sequence analysis and biochemical testing further revealed that several of the non-cooperative Paired-like factors had residues at two HD positions (28 and 43) that were not conducive to P3 site cooperativity⁹. However, these two rules were insufficient to accurately classify all Paired-like HDs as either cooperative or non-cooperative. A further limitation of the computational pipeline was that it treated cooperativity as a binary and did not capture the wide range of cooperativities seen in our biochemical validations⁹.

In this study, we used comparative quantitative DNA binding assays to assess the degree of cooperativity of 15 of the 36 Paired-like HDs on P3 sites. By combining sequence conservation, biochemical analysis, and structural modeling, we identified key residues within the HD that are critical, permissive, and inhibitory to cooperativity. We validated these residues via amino acid swaps to convert non-cooperative Paired-like HDs into cooperative proteins and vice versa. We then expanded our understanding of these mechanisms by examining sequence constraint differences between Paired-like factors and the related ANTP factors that are unable to bind cooperatively to the P3 site. This analysis revealed additional residue changes that selectively disrupt cooperativity, including residues that do not make direct protein-protein contacts but influence cooperativity through indirect mechanisms. In total, we defined 11 rules that either are critical for or contribute to cooperative DNA binding. We apply these rules to both explain why the ANTP class of HDs fail to cooperatively bind P3 sites and determine which Paired-like HDs can cooperatively bind P3 sites. Lastly, we identified 38 missense variants across 10 Paired-like proteins from HGMD that we anticipated would impact cooperativity. We tested 12 of these variants and found that all but one selectively impacted cooperativity but maintained monomer DNA binding. These findings reveal the importance of cooperativity in increasing DNA binding specificity and provide insights into how residues that do not contact DNA impact specificity. Moreover, these studies highlight how missense variants associated with at least 15 disease states can selectively disrupt cooperative DNA binding, providing a mechanistic understanding of how these disease variants impact HD function.

Results

Broad variation in cooperativity levels across the Paired-like class

While the Paired-like and ANTP subfamilies of HDs bind highly similar TAAT monomer DNA sites^1–3, HT-SELEX and biochemical studies found that some Paired-like HDs cooperatively bind TAAT – 3 N – ATTA P3 sites^2,9–13. However, the prevalence and degree of cooperative binding to the P3 site across the 36 member Paired-like class is unknown. To better define DNA binding cooperativity across the Paired-like class, we sampled 15 human Paired-like factors (Fig. 1A; highlighted in purple text) and performed quantitative electrophoretic mobility shift assays (EMSAs) using purified proteins containing their HDs and a P3 DNA probe. To select Paired-like factors for this analysis, we used three criteria: First, we created a dendrogram based on residue similarity across the entire 60 amino acid HD to identify factors from different subgroups. Second, Paired-like factors can be classified based on the identity of the 50th amino acid in the HD. K50 HDs include members of the CRX/OTX, PITX, and GSC subfamilies, whereas the other Paired-like factors and nearly all ANTP class factors encode a Q50 HD. Since K50 and Q50 HDs interact with DNA differently due to preferences for distinct nucleotides flanking the TAAT site²⁰, we selected representatives of both the K50 and Q50 subfamilies for comparative studies. Third, we prioritized factors from each subgroup based on the prevalence of reported disease-associated HD missense variants on the HGMD^21,22 (Fig. 1A).

A Dendrogram of the Paired-like class based on sequence similarity across the 60 amino acid HD. Proteins selected for cooperativity measurements are shown in purple text with the number of disease-associated missense variants within each HD denoted in parentheses^21,22. B Schematic describing calculation of cooperative binding (Tau) and relative monomer affinity (probe depletion) using a portion of the EMSAs for ALX4 shown in panel C and E. C ALX4 (amino acids = 209-274) was tested in triplicate at 25, 50, 100, and 200 nM on the indicated probes. **D, G** Tau cooperativity factors were calculated for every lane in which more than 5% of the probe was bound. Bars represent the average Tau for each protein and probe, and each dot represents a Tau from an independent binding reaction. Error bars represent mean +/- standard deviation. Tau was compared via a Welch’s two-sided t-test (n = 12). D ALX4 binds with high cooperativity to the P3Q site but not the P4Q site. E ALX4 was tested in duplicate at 30, 60, and 120 nM on a TAATTA monomer site. Probe depletion was calculated for each independent binding reaction. Each dot represents an independent binding reaction. The hill equation was used to fit the curve. The Ka was derived from the fitted Hill equation and represents the protein concentration at which half of the sites are bound. F OTP (amino acids = 99–164) was tested in triplicate at 18.75, 37.5, 75, and 150 nM on the indicated probes. G OTP binds with low cooperativity to the P3Q site but does not bind cooperatively to the P4Q site. H OTP was tested in duplicate at 27, 55, and 109 nM on a TAATTA monomer site. See (E) for plot details. I Paired-like HDs bind the P3 site with a large range of cooperativities. The Tau factors of 15 Paired-like (purple) and 3 ANTP HDs (green) were calculated from triplicate EMSAs (uncropped gels are shown in Supplementary Fig. 1). Tau cooperativity factors were calculated for every lane in which more than 5% of the probe was bound. Lower and upper hinges denote the first and third quartiles, respectively. The lower and upper whiskers denote the most extreme value if there are no outliers or 1.5 times the interquartile range. J Paired-like factor cooperativity on the P3 site is spacer length dependent as HDs are unable to bind with high cooperativity to the P4 site. Each dot represents an average Tau for a protein across all concentrations and replicates. K Cooperativity is independent of the relative HD binding affinity to a monomer site. Along the y-axis, each dot represents an average Tau on the P3 site for each protein across all concentrations and replicates (n = 18). Along the x-axis, each dot represents an average Ka of each protein on its respective high affinity monomer site across all concentrations and replicates. Error bars denote mean +/- standard deviation. A two-sided Kendall correlation analysis was performed, and the correlation coefficient and p-value are reported on the plot. Source data are provided as a Source Data file.

For comparative DNA binding analysis across the Paired-like HD family, we used P3 probes with preferred spacer sequences for Q50 versus K50 HDs. For Q50 proteins, we used a P3Q site (TAAT-TAG-ATTA) found in the most frequently bound 20mer in ALX4 HT-SELEX assays^2,9. For K50 proteins, we used a P3K site (TAAT-CCG-ATTA) that was preferably bound by the Prd K50 HD in SELEX assays¹². To assess if cooperative binding was spacer length dependent, we also tested each factor on a P4 site in which a single G was added to the P3 spacer of each respective probe (P4Q: TAAT-TAGG-ATTA; P4K: TAAT-CCGG-ATTA). As protein controls, we tested 3 ANTP factors (EN1, DLX2, and VENTX) that bind ‘TAAT‘ monomer sites but were not expected to cooperatively bind P3 sites⁹. For each protein and DNA probe, we tested four protein concentrations in triplicate in EMSAs, and quantified band intensities to calculate a cooperativity factor known as Tau¹². Tau is the ratio of two equilibrium equations: Kd₂ in which a dimer complex [P₂D] forms from a monomer complex [PD] and Kd1 in which a monomer complex is formed from unbound DNA [D] and unbound protein [P] (Fig. 1B, left panel). Tau thereby measures the multiplier by which the binding of the first protein facilitates the binding of a second protein. A tau greater than 1 is indicative of cooperative binding, a tau equal to 1 is independent binding, and a tau less than 1 indicates that the binding of the first protein inhibits the binding of the second.

To assess the relative binding affinity of DNA/protein interactions, we performed another set of quantitative EMSAs using a probe encoding a high affinity monomer site (TAATTA for Q50 HDs; TAATCC for K50 HDs). For each protein, we tested three protein concentrations in duplicate in EMSAs and quantified DNA probe depletion (Fig. 1B, right panel). By applying the Hill equation, we derived a Ka, which denotes the protein concentration at which half of the probe is bound. This approach allows us to assess cooperativity (P3 binding), spacer length specificity (P4 binding), and the relative DNA binding affinity (monomer binding) for each factor. For example, ALX4 bound in a highly cooperative manner on the P3 site with a tau of 175 ± 24.2, while OTP bound with little cooperativity with a tau of 3.57 ± 0.56, but neither protein bound cooperatively to the P4 site (ALX4 tau: 0.792 ± 0.445; OTP tau: 0.290 ± 0.242) (Fig. 1C, D, F, G; full gels in Supplementary Fig. 1A, I). It is important to note that ALX4 and OTP can bind the second site on the P4 probe but do so in an additive manner only after filling the first site. Moreover, both proteins bound similarly to a monomer site (Fig. 1E, H; full gels Supplementary Fig. 2B, C), demonstrating that ALX4 and OTP can bind similarly to a monomer sequence and palindromic sites spaced 4 bps apart, but their ability to bind as a cooperative dimer on palindromic sites spaced 3 bps apart (the P3 site) varies.

Through this survey of 15 Paired-like HDs, we found a wide range of P3 site cooperativities across the Paired-like class with OTX2, ISX, and PITX3 showing little to no cooperativity, whereas ALX4, PROP1, RAX2, ARX, and PHOX2B have a Tau value greater than 100 (Fig. 1I; full gels in Supplementary Fig. 1). Apart from OTX2 and PITX3, which were not highly cooperative on either site, we found that all Paired-like factors bound with a higher cooperativity to the P3 site compared to the P4 site (Fig. 1J; Supplementary Fig. 1). As expected, the ANTP factors, EN1, DLX2, and VENTX, did not bind cooperatively to the P3 site (Fig. 1I)⁹, and none of the ANTP factors preferred the P3 over the P4 site (Fig. 1J; Supplementary Fig. 1). There was variation in the relative binding affinity of the 18 HDs on their respective monomer sites, however we found no correlation between relative monomeric binding affinity and P3 site cooperativity (Fig. 1K; gels in Supplementary Fig. 1, Supplementary Fig. 2). Thus, while Paired-like factors share many highly conserved HD residues, they vary greatly in their ability to form cooperative DNA complexes on the P3 site.

Residue differences across the Paired-like class contributes to cooperativity variation

To identify residue differences that contribute to the variation in cooperative DNA binding, we first used all 36 Paired-like sequences to generate an information content logo of conserved residues across the HD (Fig. 2A, top). We then summarized the protein-protein contacts (red) and protein-DNA contacts (blue) using high resolution structural data of ALX4¹⁸ and Prd S50Q²³ bound to a P3 site (Fig. 2A, middle; structural alignment shown in Supplementary Fig. 3). Note, throughout this manuscript, we define protein-protein contacts as HD-to-HD contacts (denoted as Chain A and Chain B in our structural analysis) between Paired-like factors formed due to DNA-mediated cooperativity. We next aligned the amino acid sequences of the tested Paired-like HDs in order of descending tau values (Fig. 2A; bottom). Sequence analysis of HDs with low cooperativity (CRX, OTP, ISX, OTX2, and PITX3) revealed changes at five positions that make protein-protein contacts in at least one of the Paired-like structures: positions 2, 4, 28, 32, and 46 (bolded in red text, Fig. 2A). Here, we use amino acid substitutions and quantitative EMSAs to interrogate the impact of these residues on cooperative (Fig. 2B–K and Supplementary Fig. 4) and monomer DNA binding (Supplementary Fig. 5). Moreover, we test each substitution in silico with AlphaFold3²⁴ to determine if any large structural changes are predicted by the amino acid substitution (Supplementary Fig. 6). Our strategy is to swap key residues between cooperative and non-cooperative factors to identify those required to confer cooperativity and when possible, amino acid substitutions were tested in Paired-like factors that have disease associated variants at these positions. It is important to note that the cooperativity of Paired-like factors is highly variable (Fig. 1I). Because of this variability, axes are set to best visualize the respective tested proteins and should not be compared across subpanels.

Of the tested Paired-like factors, all have a glutamic acid at position 32 except ISX, which has a S32. In the ALX4 structure, Chain A E32 interacts with the N4 residue of Chain B (Fig. 2B), whereas in Prd S50Q, Chain B E32 and Chain A R3 form a salt bridge. Given the importance of E32 in both structures, we tested an E32S mutation within the highly cooperative ARX protein. EMSAs on the P3 site revealed that while wildtype ARX had a Tau of 240 ± 25, ARX E32S had a 24-fold decrease in Tau (10 ± 0.97; Fig. 2C; gels in Supplementary Fig. 4A). Inversely, the cooperativity of ISX increased ~16-fold with an S32E mutation (ISX wildtype: 2.34 ± 0.15; ISX S32E: 32.8 ± 8.5; Fig. 2D; gels in Supplementary Fig. 4B).

As mentioned, ALX4 E32 of Chain A interacts with N4 of Chain B, which is a residue that is not highly conserved across the Paired-like class (Fig. 2A). However, very few Paired-like factors have a glutamic acid at position 4 except for the weakly cooperative CRX and OTX2 factors from those surveyed (Fig. 2A, bold red text). In contrast to the ALX4 structure, S4 in the Prd S50Q structure forms no polar interactions and minimal non-polar interactions with the second protein²³. Despite this difference, a large negatively charged E4 residue would be expected to clash with E32 of the opposing chain and thereby inhibit cooperativity. To test this idea, we created an ALX4 N4E mutation and found that it reduced cooperativity ~4.5-fold (ALX4 wildtype: 176 ± 18.0; ALX4 N4E: 37.8 ± 4.33; Fig. 2E; gels in Supplementary Fig. 4C). Inversely, a CRX E4N mutation increased cooperativity 6-fold (CRX wildtype: 8.70 ± 2.34; CRX E4N: 52.7 ± 10.2; Fig. 2F; gels in Supplementary Fig. 4D). Thus, residue differences at positions 4 and 32 explain the low cooperativity levels of ISX, OTX2, and CRX compared to the other Paired-like factors.

While R2 is largely conserved across the Paired-like class, OTP has a lysine in this position. In the Prd S50Q structure, the R2 residue of both chains inserts into the minor groove of DNA, and R3 mediates protein-protein interactions²³. Conversely, in ALX4, the R2 residue of Chain A inserts into the minor groove of DNA, while Chain B R2 contacts Y25 and V28 of Chain A, revealing that this residue can directly contribute to cooperativity (Fig. 2B)¹⁸. While lysine is similarly sized and charged as arginine, lysine residues typically make fewer minor groove contacts²⁵ and are less frequent at protein-protein interfaces²⁶. To determine if K2 impacts cooperativity, we tested an ALX4 R2K mutation and found it decreased cooperativity more than four-fold (ALX4 wildtype: 167 ± 29.4; ALX4 R2K: 39.0 ± 5.08; Fig. 2G; gels in Supplementary Fig. 4E). Note, the cooperativity of ALX4 R2K was still two-fold higher than a published ALX4 R2A mutation¹⁸, which is consistent with lysine being better able to make protein-DNA and protein-protein contacts compared to alanine. To assess if K2 inhibits cooperativity in OTP, we compared OTP K2R to wildtype and found that this mutation increased cooperativity more than three-fold (OTP wildtype: 3.39 ± 0.804; OTP K2R: 11.7 ± 1.98; Fig. 2H; gels in Supplementary Fig. 4F).

While most Paired-like HDs have a Q46 residue, the non-cooperative PITX3 factor has an R46. In the Prd S50Q structure, Chain A Q46 forms a hydrogen bond with Chain B R3 (Fig. 2I)²³, while this residue makes DNA backbone contacts in the ALX4 structure¹⁸. In PITX factors, R46 is expected to clash with R3 due to size and electrostatic repulsion. Intriguingly, a Q46R missense variant in the highly cooperative PHOX2B factor has been associated with congenital central hypoventilation syndrome (CCHS)²⁷, consistent with this variant disrupting PHOX2B function. We tested the importance of Q46 in cooperativity using PHOX2B and found that this variant completely disrupted cooperativity (PHOX2B wildtype: 968 ± 67.4; PHOX2B Q46R: 1.27 ± 0.832; Fig. 2J; gels in Supplementary Fig. 4G). Conversely, we compared PITX3 wildtype and R46Q and found it significantly increased cooperativity, but only to a Tau of 2.33 (PITX3 wildtype: 0.157 ± 0.094; PITX3 R46Q: 2.33 ± 0.315; Fig. 2K; gels in Supplementary Fig. 4H). This finding suggests that other residues contribute to PITX3’s low cooperativity. Consistent with this idea, PITX3 has a methionine at position 28, and we previously found that a V28M mutation in ALX4 significantly reduced cooperativity due to clashing with the N-terminal Arginine Rich Motif (ARM) of the second protein⁹. Hence, we created a double PITX3 M28V;R46Q mutation and found that it increased cooperativity 6.5-fold compared to the PITX3 R46Q mutation and 100-fold compared to wildtype PITX3 (PITX3 M28V; R46Q: 15.3 ± 2.85; Fig. 2K). Interestingly, we found that the R46Q variant both increased cooperativity and reduced relative affinity to a monomer site by ~50% (Supplementary Fig. 5H). Conversely, all other variants tested in Fig. 2B–J, selectively disrupted P3 site cooperativity without large changes in monomer binding (Supplementary Fig. 5) and each of these amino acid swaps resulted in very little predicted structural impacts by AlphaFold3²⁴ (Supplementary Fig. 6). Altogether, these studies across Paired-like factors identified 5 HD positions (2, 4, 28, 32, and 46) that impact cooperative DNA binding, and we provide a summary of the role of each of these residues on cooperative DNA binding in Table 1.

Table 1.

Requirements of P3 site cooperativity identified from Fig. 2-Fig. 5 and past studies

HD position	Rule	Importance	Source	Structural Rationale
2	Must be an R	Contributory	This study¹⁸	R2 is required to either facilitate the protein-protein interface or insert into the minor groove.
3	Must be an R	Critical	¹⁸	R3 is required to either facilitate the protein-protein interface or insert into the minor groove.
4	Cannot be an E	Contributory	This study	E4 clashes with E32 of the partner protein
26	Must be a P	Critical	This study	Branched residues cannot fit in the narrow hydrophobic core between helix 1 and 2 that is required to form the protein-protein interface between the N-terminal ARM and turn between helix 1 and 2
28	Cannot be an R	Critical	^18,23	R clashes with the basic residues in the N-terminal ARM
32	Must be an E	Critical	This study	E32 facilitates hydrogen bonds or salt bridges with the N-terminal ARM
42	Must be an E	Critical	^18,19	E42 forms a salt bridge with R3
43	Cannot be an R	Critical	²³	R43 clashes where the recognition helices meet
44	Cannot be a Q	Critical	This study	Q44 would be unable to form the water mediated hydrogen bond with E42
46	Cannot be an R or K	Critical	This study	R46 or K46 clashes with R3 in the N-terminal ARM
50	Cannot be a K	Contributory	This study^12,19	Unknown; See discussion

Open in a new tab

HD position 50 impacts both DNA binding cooperativity and specificity

Sequence alignment and cooperativity measurements between the Paired-like HDs revealed that proteins with K50 HDs tend to have lower cooperativity levels compared to Q50 HDs (Supplementary Fig. 1T). Structurally, K50 HDs prefer a TAATCC sequence²⁰ as K50 forms hydrogen bonds with the guanines on the opposing strand²⁸, while Q50 typically does not form direct contacts with flanking bases²⁹. Previous work also revealed that the identity of the 50th HD residue can impact spacer length between sites, spacer nucleotide content preference, and cooperativity levels^12,30. To test the impact of having a Q50 vs K50 residue in cooperative DNA binding by Paired-like TFs, we mutated the highly cooperative ALX4 Q50 into a K50 and the weakly cooperative CRX K50 into a Q50. Since CRX has an E4 residue that inhibits cooperativity (see Fig. 2F), we also generated a CRX E4N;K50Q double mutation and tested each on P3 sites with linker sequences optimized for a Q50 (TAAT – TAG – ATTA; P3Q) or K50 (TAAT – CCG – ATTA; P3K) HD. Note, these variants are not predicted to cause large structural changes in the proteins (Supplementary Fig. 7). On the P3Q probe, wildtype CRX (Tau = 0.98 ± 0.68) and ALX4 Q50K (Tau = 2.02 ± 1.08) were unable to bind with high cooperativity compared to wildtype ALX4 (Tau = 156.9 ± 19.6) (Fig. 3A, B; full gels in Supplementary Fig. 8A). In contrast, ALX4 Q50K (Tau = 47.7 ± 4.04) bound cooperatively on the P3K probe, but with a significantly lower Tau than wildtype ALX4 (Tau = 343 ± 41.6) (Fig. 3A, B; full gels in Supplementary Fig. 8B). These findings suggest that a K50 protein is not conducive to strong cooperative binding on either P3 site; a result that is consistent with past studies^12,19. We also assessed the overall binding of each protein and probe by quantifying free probe depletion. As expected, wildtype ALX4 bound with a higher affinity compared to ALX4 Q50K or CRX on the P3Q probe (Fig. 3C; full gels in Supplementary Fig. 8A). However, we surprisingly found that wildtype ALX4 bound the P3K probe with similar affinity as CRX and ALX4 Q50K (Fig. 3C; full gels in Supplementary Fig. 8B). To determine if this increased ALX4 wildtype binding was related to specificity or cooperativity, we tested wildtype and Q50K ALX4 on high affinity Q50 (TAATTA) and K50 (TAATCC) monomer sequences. (Fig. 3D; full gels in Supplementary Fig. 8C, D). We found that ALX4 bound the TAATCC sequence with very low affinity (Fig. 3E), suggesting that ALX4’s ability to bind the P3K site is dependent on cooperativity. Conversely, due to the lack of strong cooperative binding, CRX and ALX4 Q50K were unable to maintain high DNA binding affinity for the P3Q site (Fig. 3C). These findings highlight how cooperativity allows for strong binding to a sub-optimal site.

Fig. 3 — A ALX4 (amino acids = 209-274), ALX4 Q50K (amino acids = 209-274), and CRX (amino acids = 34-99) were tested in triplicate at 50, 100, and 200 nM on the indicated P3Q and P3K probes. **B, G** Tau values were calculated from triplicate EMSAs using each lane where more than 5% of the probe was bound. Bars represent the average Tau for each protein across all tested concentrations and replicates, and each dot represents a Tau from an independent binding reaction. Error bars represent mean +/- standard deviation. Proteins compared with a one-way ANOVA (3 proteins). Pairwise comparisons were determined with a two-sided Welch’s t-test with a Holm multiple comparisons correction. B Protein effect: p = 4.4e-18 (ALX4 wildtype n = 9; ALX4 Q50K n = 6; CRX wildtype n = 9) G Protein effect: p = 1.78e-20 (n = 9) C, H Probe depletion was calculated for each binding reaction and compared via a two-way ANOVA. The effect of the protein variable is reported (3 proteins; 3 concentrations; n = 3). Error bars denote mean +/- standard deviation. D ALX4 wildtype and ALX4 Q50K (amino acids = 209–274) were tested in duplicate at 25, 50, and 100 nM on a TAATTA and TAATCC monomer site. **E, K** Probe depletion was calculated for each independent binding reaction and each dot represents an independent binding reaction. Data was fit with a Hill equation. F CRX wildtype, CRX K50Q, and CRX E4N; K50Q (amino acids = 34–99) were tested in triplicate at 50, 100, and 200 nM on the P3Q and P3K probes. J CRX wildtype, CRX K50Q, and CRX E4N; K50Q were tested in duplicate at 25, 50, and 100 nM on the TAATTA and TAATCC monomer probes. Source data are provided as a Source Data file.

Consistent with the Q50 residue being more compatible with cooperative DNA binding in ALX4, the CRX K50Q (P3Q Tau = 54.6 ± 2.72; P3K Tau =;340 ± 161) and CRX E4N;K50Q (P3Q Tau = 112 ± 12.1; P3K Tau = 395 ± 90.1) proteins displayed at least a 30-fold increase in cooperativity on both P3 probes compared to wildtype CRX (P3Q Tau = 0.936 ± 0.488; P3K Tau = 8.94 ± 1.78) (Fig. 3F, G; full gels in Supplementary Fig. 8F, G). This cooperative behavior enabled CRX K50Q and CRX E4N;K50Q to bind similar quantities of the P3Q and P3K probes (Fig. 3H), whereas both proteins had minimal binding on the TAATCC K50 monomer site compared to the TAATTA Q50 monomer site (Fig. 3J, K; full gels in Supplementary Fig. 8G, H). Taken together, these findings suggest that K50 residues are not conducive to strong cooperative binding and that the ability to bind DNA cooperatively can enable a protein to bind weaker half-sites with relatively high affinity.

Sequence constraints between ANTP and Paired-Like HDs identifies additional residues critical for cooperative DNA binding

While structural analysis offers valuable insight into which residues form protein-protein contacts, having only two dimer structures across the Paired-like family provides a limited understanding as to which residues are permissive or inhibitory for cooperativity. To identify additional residues that make contributions to cooperativity, we compared sequence constraints of the Paired-like class versus the ANTP class, which bind highly similar TAAT monomer sites^2,6 but fail to cooperatively bind P3 sites (Fig. 1I)⁹. In theory, differences between these classes could reveal residues that are conducive (Paired-like) or inhibitory (ANTP class) to P3 site cooperativity, while such differences are expected to maintain monomer site binding. Based on past classifications^6,31, we included 102 ANTP HDs and 36 Paired-like HDs in our analysis and performed a multiple sequence alignment of the 60 amino acid HD for all 138 factors. We converted this alignment to a similarity distance matrix and calculated the Euclidean distance between each factor. The HDs of all 36 Paired-like factors cluster together and show higher similarities to each other compared to the ANTP class (Fig. 4A–C). To systematically identify differences between classes, we created a differential bit map by subtracting the relative content information of the ANTP class from the Paired-like class (Fig. 4D). In this plot, positive bit information correlates with sequence constraint of the Paired-like class while negative bit information correlates with sequence constraint of the ANTP class. We found several areas of interest based on this analysis, including residue differences in the recognition helix as well as large differences in the turn between helix 1 and 2. As proof of concept, the differential bit analysis revealed that E32 is preferentially constrained in the Paired-like class, which is consistent with our findings on its importance in mediating protein-protein interactions that contribute to cooperative binding (Fig. 2B, D; highlighted in gray in Fig. 4D). In contrast, ANTP HDs largely have a hydrophobic residue at position 32 (Fig. 4C, D). Similarly, prior studies on Prd S50Q found that an arginine at positions 28 and 43, which are the preferred residues in ANTP HDs (highlighted in gray in Fig. 4D), disrupt cooperativity²³.

Fig. 4 — A The 60 amino acid HDs of the Paired-like class (purple) cluster together compared to the HDs of the ANTP class (green). Heatmap shows the Euclidian distances derived from the sequence alignments of the 60 amino acid DNA binding domains of 102 ANTP factors and 36 Paired-like factors, in which the smaller Euclidean distances correspond with the highest similarity between proteins. **B, C** Information bit map of the DNA binding domains of Paired-like (B) and ANTP (C) factors. Residues are colored based on amino acid properties as described in Fig. 2A. D Sequence constraint changes between the Paired-like and ANTP classes. Positive bit information correlates with sequence constraint of the Paired-like class. Negative bit information correlates with sequence constraint of the ANTP class. 3 HD positions with differential sequence constraint have been previously confirmed to alter cooperativity (denoted with gray shading), and 3 HD positions were selected for further testing (denoted with blue shading). E PYMOL schematic of the protein-protein contacts between the recognition helices and N-terminal ARMs in the Prd S50Q crystal structure (1FJL). Chain A is shown in blue and Chain B in purple. Direct protein-protein contacts denoted with black dashed lines, and water mediated contacts denoted with gray dashed lines. **F, G, I** Tau cooperativity factors were calculated from triplicate EMSAs for every lane in which more than 5% of the probe was bound. Bars represent the average Tau for each protein across all concentrations and replicates, and each dot represents a Tau from an independent binding reaction. Error bars represent mean +/- standard deviation. F Tau was compared via two-sided Welch’s t-test (n = 12). G Tau was compared via a one-way ANOVA with a Holm multiple comparisons correction (n = 9). Tau was not calculated for PHOX2B R44G due to the lack of DNA binding. H, J Protein binding to a high affinity monomer site was measured via quantitative EMSAs (gels shown in Supplementary Fig. 9). Probe depletion was calculated for each binding reaction. Each dot represents an independent binding reaction. Curves were fit with the Hill equation, and Ka values were calculated for each binding reaction in which protein was added. Printed values on plot denote Ka mean +/- standard deviation for each protein. I Tau was compared with a two-sided Welch’s t-test (PHOX2B wildtype n = 9; PHOX2B Q46K n = 12). Note: one PHOX2B wildtype lane was excluded from analysis due to a bubble in the gel. **F–J** Uncropped gels used for quantitation are shown in Supplementary Fig. 9. Source data are provided as a Source Data file.

The role of HD positions 44 and 46 in cooperativity

Two obvious differences between Paired-like and ANTP HDs occur at positions 44 and 46. At position 44, 30 of 36 Paired-like factors have an R44, whereas 99 of 102 ANTP factors have a Q44. At position 46, 29 of 36 Paired-like factors encode a Q46, whereas 94 of 102 ANTP factors have a K46. Importantly, neither an R44Q nor a Q46K Paired-like to ANTP swap are predicted to cause large structural effects (Supplementary Fig. 10). However, in the Paired S50Q structure, R44 makes water mediated contacts with either E42 or Q46 (Fig. 4E), suggesting that these swaps may impact cooperativity²³. Most water molecules could not be modeled in the ALX4 structure due to insufficient resolution, so it is unknown whether these interactions are consistent across both structures. Intriguingly, clinical genetic studies identified R44Q missense variants associated with disease in both ARX and PHOX2B^27,32, suggesting that this Paired-like to ANTP class residue swap sufficiently disrupts HD function to cause pathogenesis. To test how these variants impact DNA cooperativity, we performed comparative EMSAs with ARX wildtype (R44) and the ARX R44Q variant associated with intellectual disability and epilepsy³². These studies revealed that R44Q reduced cooperativity over 20-fold (ARX WT: 176.9 ± 35.0 vs ARX R44Q: 10.9 ± 0.7; Fig. 4F; gels in Supplementary Fig. 9A) with minimal changes to overall monomer affinity (Supplementary Fig. 9B). We similarly found that PHOX2B R44Q, which is associated with CCHS²⁷, strongly decreased cooperativity (PHOX2B wildtype: 996 ± 138; PHOX2B R44Q: 11.9 ± 1.94; Fig. 4G; gels in Supplementary Fig. 9C). Interestingly, a distinct PHOX2B missense variant at this position (R44G) was found to be associated with neuroblastoma with Hirschsprung disease³³, and this variant was previously found to decrease binding to monomer DNA sites³⁴. We similarly found that PHOX2B R44G strongly decreased DNA binding to the monomer site, whereas PHOX2B R44Q maintained similar DNA binding affinity as wildtype PHOX2B (Fig. 4H; gels in Supplementary Fig. 9D). Thus, disease variants that alter a Paired-like residue to an ANTP class residue can selectively disrupt cooperative DNA binding without dramatically impacting monomer binding, whereas a different missense variant at the same position can impact overall DNA binding.

We similarly assessed the effect of a Q46K Paired-like to ANTP swap on cooperativity by comparing the DNA binding activities of PHOX2B wildtype to PHOX2B Q46K. These studies revealed that the Q46K variant fully disrupted cooperativity (PHOX2B wildtype: 904.9 ± 126.9; PHOX2B Q46K: 0.9 ± 0.15; Fig. 4I; gels in Supplementary Fig. 9E), while slightly increasing binding affinity to the monomer site (Fig. 4J; gels in Supplementary Fig. 9F). These data are consistent with our previous finding that PHOX2B Q46R completely disrupts cooperativity (Fig. 2J), demonstrating that a positively charged residue at this position is not conducive to cooperativity, most likely due to clashing with R3 (Fig. 4E; Table 1).

The relationship between HD helical spacing and P3 site cooperativity

One of the most consistent differences between Paired-like and ANTP factors occurs at position 26, as all Paired-like factors have a proline, whereas 101 of 102 ANTP factors have a branched hydrophobic residue. Moreover, a P26L missense variant in ARX has been associated with neurological disease³⁵, suggesting this Paired-like to ANTP HD swap disrupts protein function. Analysis of the Paired-like dimer structures reveals that P26 sits near five residues that mediate protein-protein contacts that contribute to cooperative DNA binding. However, P26 does not make any direct interactions (Fig. 2A). Comparative EMSA analysis between ARX P26 and ARX P26L revealed an ~10-fold decrease in cooperativity (Fig. 5A; gels in Supplementary Fig. 11A) and no change to relative monomer binding Supplementary Fig. 11B). However, the mechanism behind this loss of cooperativity is unclear. Although L26 is accommodated and contributes to the hydrophobic core in structures of ANTP members (Fig. 5B; Supplementary Fig. 11D), alignment of the structure of the Paired-like factor, ALX4, and the ANTP factor, En, revealed that the rotamer configuration of the leucine found in En is expected to clash with position 30 in helix 2 in the ALX4 structure (Fig. 5C). To understand how a leucine fits within the hydrophobic core of ANTP factors but not Paired-like factors, we used published structures to compare the pairwise intrachain alpha carbon distances of seven Paired-like proteins and five ANTP proteins. To systematically compare distances between helices, we selected four inner helical pairs for each helix-helix combination. We found that all four helix 1 and 2 pairings were ~ 2 angstroms closer in proximity in the Paired-like structures compared to the ANTP structures, whereas there was no difference in proximity between residue pairs between helix 1 and 3 or helix 2 and 3 (Fig. 5D; Supplementary Fig. 12-13). Interestingly, the average difference of the pairwise comparisons reveals that this alteration in spacing appears to begin at position 26 (Fig. 5E). To investigate the influence of position 26 on this altered helical spacing, we compared the alpha carbon distances of AlphaFold3 predictions of the wildtype Paired-like and ANTP factors used in Fig. 5D, E and their corresponding position 26 swaps (individual comparisons are shown in Supplementary Fig. 14). This analysis suggests that a P26L variant alone in the Paired-like factors is sufficient to increase the predicted spacing between helix 1 and helix 2, and conversely that a L26P variant in the ANTP factors is sufficient to decrease this spacing (Fig. 5F). It is important to note that the predicted spacing differences were less in the single variant predictions (Fig. 5F) compared to the real change seen in the wildtype structures (Fig. 5E), which suggests that additional residues may contribute to this change. Nevertheless, this analysis suggests that position 26 is a large driver of this change in helical spacing across these HD subclasses. Further, this analysis provides insight into how a leucine can spatially fit in the ANTP hydrophobic core and not the Paired-like hydrophobic core.

Fig. 5 — A ARX P26L causes an approximately 10-fold decrease in cooperativity. Tau cooperativity factors were calculated from triplicate EMSAs using every lane in which more than 5% of the probe was bound. Bars represent the average Tau for each protein, and each dot represents a Tau from an independent binding reaction. Error bars represent mean +/- standard deviation. Tau was compared via two-sided Welch’s t-test (n = 12). Gels are shown in Supplementary Fig. 11. B PYMOL schematic showing a lack of steric clashing of L26 in Engrailed (En) (1HDD). C PYMOL schematic of steric overlap of L26 in ALX4 showing L26 clashes with A30. ALX4 is in purple and Engrailed (En) is in green. D Pairwise alpha carbon chain distances of specific residue pairings (listed at top of subpanel) across helical sets. Each datapoint represents a distance from a solved crystal structure (Paired-like factor: n = 7; ANTP factor: n = 5). Lower and upper hinges denote the first and third quartiles, respectively. The lower and upper whiskers denote the most extreme value if there are no outliers or 1.5 times the interquartile range. Pairwise distances of inner helical pairs were compared with two-sided t-tests with a Holm multiple comparisons correction. Distances for each pairing in each structure are shown in Supplementary Fig. 13. E Average pairwise alpha carbon chain distance differences across 7 Paired-like proteins and 5 ANTP proteins. Heatmap colors correspond to average distance differences in angstroms. Pairwise distances for each protein are shown in Supplementary Fig. 12. F Average pairwise alpha carbon distances of 4 AlphaFold3 predictions of Paired-like wildtype structures compared to their corresponding P26L structures (left). Average pairwise alpha carbon distances of 5 AlphaFold3 predictions of ANTP wildtype structures compared to their corresponding L26P structures (right). Heatmap colors correspond to the average distance differences in angstroms. Individual AlphaFold3 predictions are shown in Supplementary Fig. 14. G Steric overlap of residues within the dimer complex for the wildtype ALX4 dimer; the ALX4 dimer modeled with the helical spacing of the ANTP factor, En; and the En protein docked on the P3 site. Heatmap color shows the sum of the van der waals radii overlaps that were greater than the set 0.4 angstroms tolerance for each residue. The presence of steric clash is denoted by the border of the heatmap tile. H Number of residues that conflict with P3 site cooperativity based on determined rules (Table 1). Bar transparency denotes if conflicting rule is contributory or critical. Bar color denotes HD subclass. Conflicts for each factor are detailed in Supplementary Data 1. Source data are provided as a Source Data file.

To address how helical spacing influences cooperativity, we calculated the pairwise van der waals radii overlaps between the two proteins of three distinct dimer complexes on the P3 site: (1) the wildtype ALX4 dimer; (2) the ALX4 dimer modeled with the wider helical spacing found in the ANTP class; and (3) the Engrailed ANTP factor. These models revealed that the increased helical distance between helix 1 and 2 in the ANTP factors shifts the position of the turn between helix 1 and 2 and alters the energy favored rotamer conformations of residues in the turn between helix 1 and helix 2 (Supplementary Fig. 15). Cumulatively, these changes cause a significant increase in van der waals radii overlap at non-bonded residues at the protein-protein interfaces between the N-terminal ARM and turn between helix 1 and helix 2 (Fig. 5G; Supplementary Fig. 15), and these clashes would not be conducive to cooperative binding.

Defining rules of HD cooperativity on P3 sites

Based on the comparative studies between cooperative and non-cooperative Paired-like HDs (Figs. 2, 3) and the ANTP class of HDs (Figs. 4, 5), we generated 11 rules that define cooperative binding to P3 sites (Table 1). Eight of these rules are necessary for P3 site cooperativity (denoted as critical), whereas three rules influence the degree of cooperativity, but are not required for cooperativity (denoted as contributory). Applying as few as two of the critical rules to the ANTP class of HDs reveals differences that are sufficient to explain why ANTP factors fail to cooperatively bind P3 sites. First, 99 of the 102 ANTP factors have a Q44, which drastically decreases cooperativity when inserted into a cooperative Paired-like HD (Fig. 4F, G). Second, many ANTP HDs, including the 3 ANTP factors that have the required R44 (PDX1, EVX1, and EVX2), encode a positively charged residue at the 28th position, which has been shown to disrupt cooperativity on the P3 site^18,23, likely due to the positively charged side chain clashing with the N-terminal ARM. Applying all 11 rules across the ANTP family revealed that it would take a minimum of four residue changes to make an ANTP factor cooperative on P3 sites (Fig. 5H; Supplementary Data 1). Of the 36 Paired-like HDs, we identified 16 factors that break one or more rule (Fig. 5H). Seven of these Paired-like factors only conflict with contributory rules but follow all critical requirements (Fig. 5H). This analysis is consistent with our findings that GSC, OTX2, and CRX exhibit relatively low levels of cooperativity as each break at least one contributory rule (Fig. 1I; Fig. 5H).

Next, we expanded this analysis by applying the rules across all 201 human genes that encode a single 60 amino acid HD, and found that a total of 23 (20 of which are Paired-like HDs) do not break any of the 11 rules and would be expected to bind cooperatively to the P3 site, while 8 factors (7 of which are Paired-like HDs) break only contributory rules and would be expected to bind with low levels of cooperativity (Supplementary Fig. 16). The additional expected cooperative factors are predominantly from the PAX subfamily, which is consistent with past studies that have shown that the HD of the Drosophila ortholog of PAX3 and PAX7, Prd, can bind cooperatively to the P3 site¹². Based on our rule set, the atypical HD, SEBOX, is also expected to bind cooperatively to the P3 site. This factor has been previously classified as a Paired-like factor³¹ but did not cluster with the Paired-like family in our analysis due to its divergent recognition helix. However, it is important to note that many HDs outside of the ANTP and Paired-like subfamilies have additional DNA binding domains that may prevent the HD from binding cooperatively with itself.

Paired-like factor pathogenic variants impact P3 site cooperativity

We next applied these rules to disease variants across the Paired-like HD family to identify variants that impact cooperative DNA binding. Analysis of the HGMD (version 2025.2) identified 189 HD missense variants across the cooperative Paired-like factors that are associated with disease (^21,22; Fig. 6A; Supplementary Data 2). Collectively, 53 of the 189 identified disease variants impact a HD position that contributes to or is critical for cooperative DNA binding. However, some of the positions contain several different disease-associated residues with distinct biochemical properties (Fig. 6A). For example, we found that the PHOX2B R44Q variant disrupts cooperativity (Fig. 4G), whereas a PHOX2B R44G variant at this position likely destabilizes the third helix and thereby decreases DNA binding affinity (Fig. 4H). In addition to these PHOX2B disease variants, we already tested and confirmed that PHOX2B Q46R (Fig. 4I), ARX P26L (Fig. 5A) and ARX R44Q (Fig. 4F) decrease cooperativity. To further validate these hypotheses, we selected 9 additional disease variants for testing using AlphaFold3²⁴ to determine if the variants are expected to cause large structural changes (Supplementary Fig. 17) and using quantitative EMSAs on their P3 and monomer sites to assess DNA binding cooperativity and affinity.

Fig. 6 — A Pathogenic variants from HGMD^21,22 were organized by DNA binding domain position. Biochemical consequence of each pathogenic variant was assessed based on compliance to the P3 site cooperativity rules (Table 1) as well as DNA binding domain position and altered residue property. Variants expected to disrupt DNA binding are denoted in blue and variants expected to impact cooperative binding are denoted in pink. Shading denotes whether the variants’ functional consequence has been confirmed biochemically. 14 variants were tested in biochemical assays in this study. 41 variants that are hypothesized to influence cooperativity or DNA binding were tested in prior studies (see Supplementary Data 2). Contact grid shows which DNA binding domain positions are involved in protein-protein interactions (pink), DNA backbone interactions (light blue), and DNA base interactions (dark blue). **B, F** Tau cooperativity factors were calculated for every lane in which more than 5% of the probe was bound in quantitative EMSAs. Bars represent the average Tau for each protein, and each dot represents a Tau from an independent binding reaction. Error bars represent mean +/- standard deviation. B Tau was compared with a one-way ANOVA (3 proteins; n = 9; p = 1.81e-12). Pairwise two-sided Welch’s t-tests were performed with a Holm correction for multiple comparisons. C Tau was compared with a one-way ANOVA (3 proteins; n = 9; p = 1.37e-12). Pairwise two-sided Welch’s t-tests were performed with a Holm correction for multiple comparisons. D Tau was compared with a two-sided Welch’s t-test (n = 12). E Tau was compared with a one-way ANOVA (3 proteins; n = 9; p = 1.07e-22). Pairwise two-sided Welch’s t-tests were performed with a Holm correction for multiple comparisons. F Tau was compared with a two-sided Welch’s t-test (n = 12). Uncropped gels are shown in Supplementary Fig. 18. Source data are provided as a Source Data file.

In total, 3 positions in the N-terminal ARM are critical for or contribute to cooperativity (Table 1). Of these positions, 6 missense variants in R2 (contributory) and 12 missense variants in R3 (critical) have been associated with disease states. In ALX4, we previously used alanine mutagenesis to show that having an arginine at both position 2 and 3 is required for cooperative DNA binding due to their contributions to protein-protein interactions. In contrast, a single arginine at one of these positions was sufficient to maintain DNA binding affinity as either can insert into the minor groove and contact DNA. We also found that R3A disrupted cooperativity to a greater degree compared to R2A in ALX4¹⁸, consistent with R2 contributing to cooperative binding, whereas R3 is critical to cooperative binding (Table 1). While alanine was not found among the disease associated missense variants in Paired-like HDs at positions 2 and 3, leucine, which is larger, but biochemically similar to alanine, was associated with disease states in PHOX2B (R2L³⁶ and R3L³³) and ARX (R3L³⁷) (Supplementary Data 2). Consistent with our prior ALX4 data, we found that both PHOX2B leucine variants decreased cooperativity with R3L doing so to a greater extent than R2L (PHOX2B wildtype: 894 ± 242; PHOX2B R2L: 167 ± 24.6; PHOX2B R3L: 5.40 ± 0.500; Fig. 6B; gels in Supplementary Fig. 18A). Moreover, like ALX4 R2A and R3A, each of the PHOX2B variants caused only minor changes to relative monomer binding affinity (Supplementary Fig. 18B).

We performed a similar analysis on variants in CRX, which has been reported to have R2Q, R3Q, and R3W variants associated with cone rod dystrophy^38,39. AlphaFold3 predicts that CRX R3Q would fail to facilitate the salt bridge with E42 at the protein-protein interface and suggests this residue would not maintain all hydrogen bonds in the minor groove of DNA (Supplementary Fig. 19E, F). Because of this, we expected that a CRX R3Q variant would use its R2 residue to bond with the minor groove of DNA, resulting in a loss of the protein-protein interface between the N-terminal ARM and turn between helix 1 and helix 2. However, this residue is not expected to sterically clash with other residues at either interface (Supplementary Fig. 19E, F). This contrasts with CRX R3W which is not only predicted to be unable to facilitate protein-protein or protein-DNA contacts but also predicted to force R2 into a position where it is unable to form these critical DNA contacts (Supplementary Fig. 19G, H). Hence, we expect CRX R3W to cause large disruptions to DNA binding. Consistent with these in silico modeling predictions, we found that CRX R3Q disrupts cooperativity (Fig. 6C; gels in Supplementary Fig. 18C) but has similar relative monomer affinity compared to wildtype (Supplementary Fig. 18D), whereas CRX R3W strongly disrupts DNA binding affinity (Supplementary Fig. 18K).

Based on past mutagenesis studies in ALX4¹⁸, we expected the CRX R2Q variant to behave similarly as CRX R3Q, ALX4 R2A¹⁸, and PHOX2B R2L (Fig. 6B) and disrupt cooperativity while maintaining monomer DNA binding. In contrast, AlphaFold3 predicted a R2Q residue change would disrupt protein-DNA interactions in the minor groove and thereby decrease overall DNA binding affinity (Supplementary Fig. 19C, D). Contrary to both predictions, CRX R2Q had no impact on cooperativity or monomer affinity (Fig. 6C; gels in Supplementary Fig. 18C). It is important to note however, that the ALX4 R2A and PHOX2B R2L variants decreased cooperativity to a tau of 20¹⁸ and 167 (Fig. 6B) respectively, which are both greater than CRX wildtype’s tau. Because of this, it is possible that the R2Q variant would have had a greater impact on cooperativity in a protein with a higher wildtype cooperativity.

The 4th residue within the N-terminal ARM also contributes to cooperative DNA binding. To date, there is only one disease associated variant in the 4th position, CRX E4K, which is associated with Leber congenital amaurosis⁴⁰. We previously showed that altering the CRX E4 residue into an asparagine (E4N) resulted in higher cooperativity than wildtype CRX (Fig. 2F) as this residue change eliminates charge clashes between E4 and E32 and possibly facilitates polar bonds between E32 and N4, like those in the ALX4 dimer structure (Fig. 2B). As a lysine is positively charged and not expected to clash with the negatively charged glutamic acid, we hypothesize that this disease variant will increase cooperative binding (Table 1). Consistent with this idea, CRX E4K bound with an 18-fold increase in cooperativity relative to wildtype CRX (CRX wildtype: 9.04 ± 2.79; CRX E4K: 167 ± 51.1; (Fig. 6D; gels in Supplementary Fig. 18E), with little change in relative monomer affinity (Supplementary Fig. 18D). The increased cooperativity may act similarly to a second Leber congenital amaurosis variant, CRX K50N, which was previously shown to increase cooperativity on the P3 site, and thereby cause ectopic binding to Q50 genomic sites¹⁹.

Lastly, we found that HD positions 42, 43, 44, and 46 within the third helix are critical for cooperative DNA binding and position 50 is contributory. In total, 20 different Paired-like HD missense variants at these positions have been associated with disease. Of these, we tested CCHS variants, PHOX2B E42Q⁴¹ and PHOX2B A43E⁴², and found both decrease cooperativity (Fig. 6E; gels in Supplementary Fig. 18F) with only minor changes in affinity (Supplementary Fig. 18G). The PHOX2B E42Q variant likely disrupts the R3-E42 salt bridge^18,23, which is consistent with prior work showing that ALX4 E42A fully disrupts cooperativity¹⁸ as well as a study showing that CRX E42A decreased dimeric binding in in vitro DNA binding assays and genomic assays in the mouse retina¹⁹. PHOX2B A43E most likely clashes at the interface between the two recognition helices, similarly to what was seen in the ALX4 A43V and ALX4 A43D variants⁹. Moreover, we confirmed that the disease variant associated with autism spectrum disorder, ARX A43V³², also caused a 7.5-fold decrease in cooperativity (Fig. 6F; gels in Supplementary Fig. 18H) without altering relative monomer affinity (Supplementary Fig. 18I).

In summary, we identified 38 human disease variants expected to alter cooperativity (Fig. 6A; Supplementary Data 2). After biochemical testing, we found that 11 of 12 of these variants selectively disrupted cooperativity without large impacts on DNA binding affinity. This work complements two recent studies that identified four human disease variants that alter cooperativity^18,19. Collectively, these findings highlight how Paired-like HD variants that selectively disrupt cooperative DNA binding can sufficiently alter protein function to cause disease.

Discussion

Our findings in this study provide insight to two questions: How do Paired-like HDs gain target specificity to regulate diverse developmental processes, and how do missense alleles impact HD function to cause disease? Here, we found that Paired-like HDs vary greatly in their ability to cooperatively bind the palindromic P3 DNA site. By combining sequence constraint analysis within the Paired-like family and against the non-cooperative ANTP HDs, protein-protein contact information from structural studies, and quantitative DNA binding measurements, we defined 11 rules that are critical for or contribute to the dramatic differences in cooperative DNA binding between members of the Paired-like HDs. We applied these rules to: (1) Predict the cooperative behavior of the remaining Paired-like HDs; (2) Explain why members of the ANTP subfamily fail to cooperatively bind the P3 site; (3) Assess the functional impacts of human disease variants across Paired-like HDs. From the variant allele analysis, we identified 38 disease associated variants that we hypothesized would alter cooperative DNA binding and performed validation studies to show that 11 of the 12 tested variants selectively altered cooperativity without greatly impacting DNA binding affinity. Collectively, these studies have two important implications: First, they show how differences in HD residues at positions not normally associated with DNA binding specificity can contribute to cooperative DNA binding and thereby help explain the different binding behaviors as well as the differential target specificity found in the Paired-like factors. Second, they reveal how disease associated missense variants in Paired-like factors can be stratified into different groups based upon their impact on cooperative DNA binding versus overall DNA binding affinity.

Sequence constraint identifies HD residues that contribute to cooperative DNA binding and specificity

The way in which HD TFs gain specificity to bind distinct genomic targets using highly similar DNA binding domains has been a long-standing question in the field. Here, we took advantage of the large number of human HDs to compare sequence constraints across the Paired-like class, a subfamily of HDs in which many, but not all, members cooperatively bind the P3 site, and the ANTP family members that do not cooperatively bind the P3 site. Through this analysis, we found that select Paired-like HDs’ cooperative behavior on the P3 site requires residues that do not contact DNA but contribute to complex formation via direct protein-protein interactions or indirect mechanisms such as defining helical spacing. Importantly, many of these residues are largely specific to the Paired-like class and not conserved in the non-cooperative ANTP class (Fig. 4D). Further, these residues are not conserved across all Paired-like members and vary in factors that are weakly or non-cooperative (Fig. 2A). This contrasts with many of the residues involved in direct protein-DNA contacts, such as the R5 and N51 residues, which are largely conserved across all HDs. Thus, while Paired-like and ANTP class HDs bind highly similar TAAT monomer sites using the same HD residues, residue differences between Paired-like and ANTP factors at non-DNA contacting positions, especially within the turn between Helix 1 and 2 (positions 26, 28, and 32) and at the beginning of the 3rd helix (positions 42, 43, 44, and 46) increase HD DNA binding specificity by contributing to and stabilizing the formation of cooperative dimer complexes on the P3 site.

Of the 11 rules found to contribute to Paired-like HD cooperativity on P3 sites, HD positions 2, 3, 28, 42, and 43 were initially described in prior structural studies on ALX4 and Prd S50Q^18,23. Here, we validated those studies as well as further identified and elucidated the impacts of HD positions 4, 26, 32, 44, 46, and 50 on cooperative complex formation. Four of these residues (4, 32, 46, and 50) were identified by direct comparative analysis of cooperative versus non-cooperative Paired-like factors. For example, our data using ALX4 and CRX proteins that swap only the 50th amino acid reveals that a Q50 residue is conducive to cooperativity, and this cooperativity enhances DNA binding specificity by enabling proteins to bind HD sites with suboptimal flanking regions (Fig. 3). While the structural mechanism behind the decreased cooperativity of K50 HDs is unclear, we can offer two possibilities: (1) Q50 typically interacts with DNA through weaker van der waals or water mediated interactions compared to K50, which typically forms direct hydrogen bonds with the guanines of the opposing strand on a TAATCC site²⁸. Hence, it is possible that the stronger bonds with the central nucleotides of the spacer sequence between the HD sites cannot be displaced and thereby prevents a second K50 protein from binding the site. (2) We previously found that the Gsx2 HD, which binds as a dimer on two TAAT sites spaced 7 bps apart, has enhanced cooperativity on DNA with an AT-rich versus a GC-rich spacer⁴³. As much of the spacer sequence is not directly contacted by Gsx2, these data suggest that DNA flexibility and/or DNA shape directly contributes to cooperativity. However, in the case of the Paired-like factors, the HD sites are much closer together and the “spacer” sequences make numerous direct interactions with HD residues. Hence, if a similar mechanism affects the K50 vs Q50 Paired-like factors, there may be tradeoffs between the best spacer sequences for optimal shape/flexibility for cooperativity and the best spacer sequences for specificity through interactions with the K50 vs Q50 residues.

By leveraging sequence constraint differences and residue preferences between the Paired-like and ANTP classes, we identified three additional residues (26, 44, and 46) that contribute to cooperative DNA binding. Of these residues, structural analysis revealed mechanistic underpinnings for the R44Q and Q46K Paired-like to ANTP-like swaps due to the disruption of direct or water-mediated protein-protein interactions. However, the reason behind the loss of cooperativity in the P26L Paired-like to ANTP-like swap was not immediately apparent in the ALX4 or the Prd S50Q dimer structures. An in-depth structural analysis revealed that helix 1 and helix 2 are closer in proximity in the Paired-like structures compared to the ANTP structures (Fig. 5D, E), and we found that this change appears to be partially driven by position 26 (Fig. 5F). We further found that this change in spacing would cause significant clashing at the protein-protein interfaces between the N-terminal ARM and turn between helix 1 and helix 2 (Fig. 5G). However, we can only speculate as to how position 26 drives this change as well as what other residues contribute to the distinct helical spacings between these families. It is possible that the proline in Paired-like factors increases the rigidity in the turn between helices, providing the required strict and tight spacing. However, it is also possible that other covarying residues contribute to the distinct helical spacing between helix 1 and helix 2. For example, a previous covariation analysis found that HDs containing a branched aliphatic residue at position 26 formed a salt bridge between positions 19 and 30 in 92% of cases. In contrast, if a HD contains a proline at position 26, as found in all Paired-like HDs, positions 19 and 30 are always hydrophobic^44,45. It is important to note that in most HDs there is a single salt bridge that connects each helical pair, and these are positions 19 and 30 between helix 1 and 2, positions 17 and 52 between helix 1 and 3, and positions 31 and 42 between helix 2 and 3⁴⁴. Of these, the salt bridge mediating residues for helix 1 and 3 as well as helix 2 and 3 are conserved between the Paired-like and ANTP factors (Fig. 4B, D), which correlates well with the conserved helical spacing across these pairings (Fig. 5D-E). However, the residues at positions 19 and 30 are distinct between the families (Fig. 4B, D), which may correlate with the observed change in helical spacing. This strong covariant relationship could also be related to either the 19-30 hydrophobic bridge or L26 being required for proper hydrophobic core formation. In the ALX4 structure, V19 and A30 insert into the hydrophobic core at the connection between helix 1 and 2 in a similar position to L26 in the Engrailed structure (Supplementary Fig. 11D, E). Future studies will be needed to investigate the influence of the 19-30 salt bridge versus hydrophobic bridge as well as increased rigidity possibly provided by P26 versus the L26 hydrophobic core contributions in the determination of helical spacing, stability, and cooperativity. As such, although the 11 rules help explain the varying abilities to bind cooperatively to the P3 site, it is likely that other HD positions contribute to cooperativity. Consistent with this idea, our rules fail to explain the large range of cooperative behaviors in the 20 Paired-like factors that follow all criteria. Hence, computational modeling that incorporates amino acid sequence, direct coupling analyses, and evolutionary constraint will be needed to further elucidate the relationship between amino acid sequence and cooperativity.

Elucidating the impact of Paired-like HD cooperativity on protein function and disease variant mechanisms

By applying our cooperativity rules to human variants found in the HGMD^21,22, we identified 38 variants expected to impact cooperativity, whereas an additional 75 variants would likely impact overall DNA binding affinity (Fig. 6A; Supplementary Data 2). In some cases, cooperativity helps clarify potential genotype to phenotype relationships of human disease variants. For example, ARX variants have been associated with epilepsy, lissencephaly, intellectual disability, corpus collosum abnormalities, and autism^21,22. Interestingly, there appears to be a loose correlation between variant impact on protein function and disease severity. All four ARX variants that occur at position 5, a position known to be critical for DNA binding and to strongly disrupt DNA binding when altered in ARX^46–48, cause severe lissencephaly or corpus collosum abnormalities^49–52, whereas variants that are expected to selectively disrupt cooperativity such as ARX R3L and ARX P26S as well as variants we confirmed disrupt cooperativity, ARX P26L (Fig. 5A), ARX A43V (Fig. 6F), and ARX R44Q (Fig. 4F) are associated with epilepsy and intellectual disability^32,35,37,53. This correlation can be seen at single DNA binding domain positions as well. For example, the ARX P26L variant that disrupts cooperativity, but has no impact on DNA binding affinity, has been associated with epilepsy, whereas ARX P26R disrupts all DNA binding^5,48 and has been associated with lissencephaly, a severe structural brain defect⁵². Altogether, these studies suggest that a variant’s impact on protein function could be derived from disease associations in cases where one factor has associations with disease states of varying severity.

A limitation of our study is that we tested protein fragments that only encode the DNA binding domain on only one or two P3 sites that were preferred by Paired-like Q50 and K50 proteins in past assays^2,9,12. However, previous work has shown that although the DNA spacer sequence content alters the magnitude of cooperative binding, it does not change the ability of ALX1, ALX4, or ARX to cooperatively bind the P3 site⁹. Despite these findings, we cannot rule out the possibility that select proteins and variants may behave differently on distinct P3 sites. Our study also fails to capture the behavior of the full-length protein in its genomic environment. However, several Paired-like factors have genomic binding datasets available and show evidence of in vivo binding to the P3 site. For example, ALX4 bound 519 P3 sites in human cranial neural crest cells^17,18; P3 sites were present in over 75% of peaks bound by Phox2a in cortical motor neurons^9,14; and P3 sites were present in over 85% of PHOX2B bound peaks in KELLY, BE2C, and CLBGA neuroblastoma cell lines^9,15,16. Hence, these genomic studies provide evidence that our biochemistry results are representative of what occurs in vivo.

Perhaps the best understood Paired-like HD factor in terms of genomic binding and transcriptional function is the weakly cooperative CRX protein. CRX is essential for photoreceptor development and genetic variants in CRX have been associated with cone-rod dystrophy, retinitis pigmentosa, macular dystrophy, and Leber congenital amaurosis^21,22. At the transcriptional level, a massive parallel reporter assay comparing monomer and P3 dimer sites showed that CRX binding to dimer sites correlates with higher transcriptional activation levels⁵⁴. Moreover, a genomic study in the mouse retina revealed that CRX variants that either increase or decrease cooperativity to P3 sites can cause disease¹⁹. For example, a CRX K50N variant associated with Leber congenital amaurosis both increased CRX cooperativity and altered DNA binding specificity, much like we found using the CRX K50Q variant (Fig. 3). Importantly, the enhanced cooperativity and specificity change led to ectopic binding at Q50 sites that are not bound by wildtype CRX, thereby compromising photoreceptor development^19,55. In contrast, a CRX E42A variant associated with cone rod dystrophy was found to disrupt cooperativity and behaved similarly as PHOX2B E42Q (Fig. 6E) and ALX4 E42A¹⁸. The cooperativity-deficient CRX E42A variant led to accelerated chromatin remodeling in early photoreceptor development due to increased K50 monomer site binding, and loss of chromatin remodeling at the end of development due to decreased K50 dimer site binding¹⁹. Notably, this study revealed that 61.27% of target genes are regulated by enhancers that contain both dimeric and monomeric sites, and these genes are selectively downregulated in the CRX E42A variant compared to CRX WT, highlighting the importance of cooperative binding in transcriptional regulation, even in factors that exhibit low levels of cooperativity¹⁹. Taken together, these studies reveal the importance of accurate cooperativity levels for the proper function of Paired-like HDs and suggest that stratifying disease variants based upon molecular defect may reveal new insights into genotype/phenotype relationships associated with disease.

Methods

Protein preparation

For each Paired-like protein tested in this study, we used a 66 amino acid region including the 60 amino acid HD and 5 amino acids N-terminal and 1 amino acid C-terminal of the HD as this truncation was previously found to be well-behaved in ALX4 biochemical, biophysical, and structural studies¹⁸. Further, residues flanking the ALX4 HD were found to form protein-protein contacts in the ALX4 structure¹⁸. Each 66 amino acid region was PCR amplified from cDNAs with Accuzyme DNA polymerase (Bioline) or KOD Xtreme Hot Start DNA polymerase (Novagen) following manufacturer’s protocols. Primers and template DNAs used to amplify each gene are detailed in Supplementary Data 3. PCR products were cut with either NdeI/XhoI or BamHI/NotI (NEB) and ligated into pET14P⁹ with T4 ligase (NEB) following manufacturer’s protocol. All sequences were confirmed via Sanger sequencing. All cDNAs are available upon request. Constructs were transformed into BL21 or C41(DE3) E.coli, grown in LB media to an optical density between 0.6 and 0.8, and induced in 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). Proteins were purified with Ni-NTA beads (Qiagen). Protein purity was assessed via SDS-PAGE with GelCode blue staining (Thermo Scientific), and gels were scanned on the Li-Cor Odyssey CLx scanner. Total protein concentrations were determined via Bradford assay (BioRad). If protein purities differed between wildtype and variant proteins, we quantified band intensities using Li-Cor Image Studio software (v1.0.9) and normalized the variant protein concentration to the wildtype.

Electrophoretic mobility shift assays

Probes were prepared by annealing a IRDye 700 nM labeled 5’ - AGCCACGCCCGCACTA - 3’ oligo to an oligo containing the binding sites of interest followed by filling in the DNA with Klenow (NEB). The sequences of the oligos containing the binding sites of interest are denoted in Supplementary Data 4. All EMSA binding reactions were incubated for 20 minutes at room temperature with 10 mM Tris, pH 7.5; 50 mM NaCl; 1 mM MgCl₂; 4% glycerol; 0.5 mM DTT; 0.5 mM EDTA; 50 µg/ml poly(dI–dC); and 200 µg/ml of BSA, 34 nM of denoted fluorescent probe, and the protein of interest. Protein concentrations are listed in figure captions for each respective experiment. Binding reactions were run for 2 h at 150 volts on a 4% polyacrylamide gel and then scanned on the Li-Cor Odyssey CLx scanner. Band intensity measurements were made in the Li-Cor Image Studio software. All uncropped and unprocessed gels are provided in the Supplementary Figs. Tau was calculated by the equation below where [P₂D] represents dimer binding, [PD] represents monomer binding, and [D] represents unbound probe as previously described¹². Tau cooperativity factors were calculated for every lane in which more than 5% of the probe was bound.

τ = \frac{4 [P_{2} D] [D]}{{[P D]}^{2}}

AlphaFold3 predictions

The 66 amino acid variant and corresponding wildtype on the P3 site was predicted with AlphaFold3²⁴. The wildtype and variant alpha carbon chains were aligned with the PYMOL (v2.5.7) align function. The root means squared deviation between the wildtype and variant structures were calculated in PYMOL. The alpha carbon distances within the proteins were measured with the get_distance command in PYMOL.

Sequence constraint analysis

We identified 228 human HDs from CISBP⁵⁶ and found DNA binding domain annotation for 227 of these in UNIPROT⁵⁷. HD classifications were performed via clustering, guided by definitions from Holland et al 2007³¹. We continued our analysis with the Paired-like and ANTP factors, which lack additional DNA binding domains. For simplicity, we only continued with factors with exactly 60 amino acid DNA binding domains. We generated multiple sequence alignments of the 102 ANTP and 36 Paired-like DNA binding domains⁵⁸, converted these alignments to a matrix of Euclidian distances⁵⁹, and then clustered with the complete linkage method and visualized with pheatmap. Bit information of DNA binding domain sequences was calculated and plotted with ggseqlogo⁶⁰. Differential bit information was calculated by subtracting the bit information of the ANTP factors from the Paired-like factors at each residue and position. Scripts used for this analysis can be found on GitHub [https://github.com/cainbn97/PairedLikeCooperativity]⁶¹.

Helical spacing analysis

Alpha carbon distances were measured with the get_distance function in PYMOL (v2.5.7). Structures used in the analysis in Figs. 5D, E were published previously^{18,23,29,43,62–66}. The ALX4 structure was modeled with the Engrailed helical spacing by segmenting the structure into the N-terminal ARM (residues 1-8), helix 1 (residues 9-25), helix 2 (residues 26-39), and helix 3 (residues 40-60); aligning each helix to the like region of the Engrailed structure; creating bonds between the segments to generate a single object; and cleaning all residues within 10 angstroms of the new bond with the clean function. Steric overlap was calculated in PYMOL by first calculating the atomic distances between each atom-atom combination of a residue-residue pairing. Any distance that was less than the sum of the pairwise atomic field radii minus 0.4 angstroms (used as tolerance) was considered a steric clash. Scripts used for this analysis can be found on GitHub [https://github.com/cainbn97/PairedLikeCooperativity]⁶¹.

Statistics and reproducibility

No statistical method was used to predetermine sample size for Fig. 1. Four concentrations were tested in triplicate (n = 12) for each protein on the P3 and P4 probe. We estimated the sample size needed to measure a two-fold change in cooperativity at a significance level of 0.05 and a power of 0.80 using the derived means and standard deviations of the wildtype proteins on the P3 probes from Fig. 1 (Supplementary Data 5). As the number of samples of the Paired-like factors needed to fulfil this criterion predominantly ranged from 2-7 samples per group, we tested 9 to 12 binding reactions for each protein and probe. The sample sizes for EMSA binding reactions within this range were based on the maximum number of replicates that could be run in parallel on our 2 gel, 15 lane gel apparatus. Samples sizes for each experiment are denoted in the figure captions and Source Data file. Cooperativity (Tau) was quantified for all binding reactions in which more than 5% of the probe was bound. All binding reactions were used for relative monomer affinity quantitation. As there is no group allocation in this study, blinding and randomization is not applicable.

For each statistical comparison, a Shapiro-Wilks test was used to assess normality. If the populations were not normal, a two-sided Wilcox rank sum test was used to compare groups. If populations were normal, a Levene’s test was used to assess the homogeneity of variance. If the Levene’s test revealed equal variance, a two-sided student t-test was performed. If the Levene’s test revealed unequal variance, a two-sided Welch’s t-test was performed. If the statistical comparison included more than two groups, we first performed a one-way ANOVA and then if significant, performed pairwise student or Welch’s t-test depending on the Levene’s test result with a Holm correction for multiple comparisons. For each test, we calculated Cohen’s d effect size. The results of the Shapiro-wilks normality tests and Levene’s test for homogeneity of variances as well as the degrees of freedom, test statistics, confidence intervals, and effect sizes for all null hypothesis testing in the manuscript can be found in the Source Data file.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information^{(6.4MB, pdf)}

Transparent Peer Review file^{(682.7KB, pdf)}

Reporting Summary^{(85.9KB, pdf)}

41467_2026_76294_MOESM4_ESM.pdf^{(103.8KB, pdf)}

Description of Additional Supplementary Files

Supplementary Data 1-5^{(96.9KB, xlsx)}

Source data

Source Data^{(4.8MB, xlsx)}

Acknowledgements

We would like to thank Richard Mann, Shaoxun Liu, and Jeffrey Steimle for their helpful comments and suggestions on an earlier version of this manuscript.

Author contributions

The scientific study was conceived and planned by B.C. and B.G. Construct synthesis and protein purification were performed by B.C., C.W., and F.R. EMSAs were performed by B.C., B.G., C.W., and F.R. All bioinformatic and structural analyses were performed by B.C. The manuscript was written by B.C. and B.G. and edited by all authors.

Peer review

Peer review information

Nature Communications thanks Raktim Mitra and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Institute of Health grants to B.G (R01-GM079428, R35-GM158075 and R01-NS124660) and a Center for Pediatric Genomics grant from CCHMC to B.G.

Data availability

All data generated in this study are found in the article and source data file. Source data are provided with this paper.

Code availability

All custom scripts can be found at GitHub[https://github.com/cainbn97/PairedLikeCooperativity]⁶¹.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Contributor Information

Brittany Cain, Email: Brittany.cain@cchmc.org.

Brian Gebelein, Email: brian.gebelein@cchmc.org.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76294-w.

References

1.Lambert, S. A. et al. The Human Transcription Factors. Cell172, 650–665 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Jolma, A. et al. DNA-binding specificities of human transcription factors. Cell152, 327–339 (2013). [DOI] [PubMed] [Google Scholar]
3.Berger, M. F. et al. Variation in Homeodomain DNA Binding Revealed by High-Resolution Analysis of Sequence Preferences. Cell133, 1266–1276 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Badis, G. et al. Diversity and complexity in DNA recognition by transcription factors. Science (1979)324, 1720–1723 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Kock, K. H. et al. DNA binding analysis of rare variants in homeodomains reveals homeodomain specificity-determining residues. Nat. Commun. 15, 3110 (2024). [DOI] [PMC free article] [PubMed]
6.Bürglin, T. R. & Affolter, M. Homeodomain proteins: an update. Chromosoma125, 497–521 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Jun, S. & Desplan, C. Cooperative interactions between paired domain and homeodomain. Development122, 2639–2650 (1996). [DOI] [PubMed] [Google Scholar]
8.Slattery, M. et al. Cofactor binding evokes latent differences in DNA binding specificity between hox proteins. Cell147, 1270–1282 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Cain, B. et al. Prediction of cooperative homeodomain DNA binding sites from high-throughput-SELEX data. Nucleic Acids Res51, 6055–6072 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Qu, S., Tucker, S. C., Zhao, Q. & Wisdom, R. Physical and genetic interactions between Alx4 and Cart1. Development369, 359–369 (1999). [DOI] [PubMed] [Google Scholar]
11.Cai, R. L. Human CART1, a paired-class homeodomain protein, activates transcription through palindromic binding sites. Biochem. Biophys. Res. Commun.250, 305–311 (1998). [DOI] [PubMed] [Google Scholar]
12.Wilson, D., Sheng, G., Lecuit, T., Dostatni, N. & Desplan, C. Cooperative dimerization of paired class homeo domains on DNA. Genes Dev.7, 2120–2134 (1993). [DOI] [PubMed] [Google Scholar]
13.Tucker, S. C. & Wisdom, R. Site-specific heterodimerization by paired class homeodomain proteins mediates selective transcriptional responses. J. Biol. Chem.274, 32325–32332 (1999). [DOI] [PubMed] [Google Scholar]
14.Mazzoni, E. O. et al. Synergistic binding of transcription factors to cell-specific enhancers programs motor neuron identity. Nat. Neurosci.16, 1219–1227 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Durbin, A. D. et al. Selective gene dependencies in MYCN-amplified neuroblastoma include the core transcriptional regulatory circuitry. Nat. Genet.50, 1240–1246 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Boeva, V. et al. Heterogeneity of neuroblastoma cell identity defined by transcriptional circuitries. Nat. Genet.49, 1408–1413 (2017). [DOI] [PubMed] [Google Scholar]
17.Kim, S. et al. DNA-guided transcription factor cooperativity shapes face and limb mesenchyme. Cell187, 692–711 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Cain, B., Yuan, Z., Thoman, E., Kovall, R. A. & Gebelein, B. The ALX4 dimer structure provides insight into how disease alleles impact function. Nat. Commun.16, 4800 (2025). [DOI] [PMC free article] [PubMed]
19.Zheng, Y., Stormo, G. D. & Chen, S. Aberrant homeodomain–DNA cooperative dimerization underlies distinct developmental defects in two dominant CRX retinopathy models. Genome Res. 10.1101/gr.279340.124 (2024) [DOI] [PMC free article] [PubMed]
20.Treisman, J., Gonczy, P., Vashishtha, M., Harris, E. & Desplan, C. A Single Amino Acid Can Determine the DNA Binding Specificity of Homeodomain Proteins. Cell59, 553–562 (1989). [DOI] [PubMed]
21.Stenson, P. D. et al. The Human Gene Mutation Database (HGMD®): optimizing its use in a clinical diagnostic or research setting. Hum. Genet.139, 1197–1207 (2020). [DOI] [PMC free article] [PubMed]
22.Stenson, P. D. et al. The Human Gene Mutation Database: towards a comprehensive repository of inherited mutation data for medical research, genetic diagnosis and next-generation sequencing studies. Hum. Genet.136, 665–677 (2017). [DOI] [PMC free article] [PubMed]
23.Wilson, D. S., Guenther, B., Desplan, C. & Kuriyan, J. High resolution crystal structure of a paired (Pax) class cooperative homeodomain dimer on DNA. Cell82, 709–719 (1995). [DOI] [PubMed] [Google Scholar]
24.Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature630, 493–500 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Rohs, R. et al. The role of DNA shape in protein-DNA recognition. Nature461, 1248–1253 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Tsai, C. J., Lin, S. L., Wolfson, H. J. & Nussinov, R. Studies of protein-protein interfaces: A statistical analysis of the hydrophobic effect. Protein Sci.6, 53–64 (1997). [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Trochet, D. et al. PHOX2B Genotype Allows for Prediction of Tumor Risk in Congenital Central Hypoventilation Syndrome. Am. J. Hum. Genet.76, 421–426 (2005). [DOI] [PMC free article] [PubMed]
28.Tucker-Kellogg, L. et al. Engrailed (Gln50→Lys) Homeodomain-DNA Complex at 1.9 Å Resolution: Structural Basis for Enhanced Affinity and Altered Specificity. Structure5, 1047–1054 (1997). [DOI] [PubMed]
29.Kissinger, C. R., Liu, B., Martin-Blanco, E., Kornberg, T. B. & Pabo, C. O. Crystal structure of an engrailed homeodomain-DNA complex at 2.8 Å resolution: A framework for understanding homeodomain-DNA interactions. Cell63, 579–590 (1990). [DOI] [PubMed] [Google Scholar]
30.Wilson, D. S., Sheng, G., Jun, S. & Desplan, C. Conservation and Diversification in Homeodomain-DNA Interactions: A Comparative Genetic Analysis. Proc. Natl. Acad. Sci. USA93, 6886–6891 (1996). [DOI] [PMC free article] [PubMed]
31.Holland, P. W. H., Booth, H. A. F. & Bruford, E. A. Classification and nomenclature of all human homeobox genes. BMC Biol. 5, (2007). [DOI] [PMC free article] [PubMed]
32.Thai, M. H. N. et al. Constraint and conservation of paired-type homeodomains predicts the clinical outcome of missense variants of uncertain significance. Hum. Mutat.41, 1407–1424 (2020). [DOI] [PubMed] [Google Scholar]
33.Trochet, D. et al. Germline Mutations of the Paired-Like Homeobox 2B (PHOX2B) Gene in Neuroblastoma. Am. J. Hum. Genet.74, 761–764 (2004). [DOI] [PMC free article] [PubMed]
34.Trochet, D. et al. Molecular consequences of PHOX2B missense, frameshift and alanine expansion mutations leading to autonomic dysfunction. Hum. Mol. Genet.14, 3697–3708 (2005). [DOI] [PubMed] [Google Scholar]
35.Strømme, P. et al. Mutations in the human ortholog of Aristaless cause X-linked mental retardation and epilepsy. Nat. Genet.30, 441–445 (2002). [DOI] [PubMed] [Google Scholar]
36.Monies, D. et al. The landscape of genetic diseases in Saudi Arabia based on the first 1000 diagnostic panels and exomes. Hum. Genet.136, 921–939 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Gong, X. et al. Generation of an isogenic gene-corrected iPSC line (OGHFUi001-A-1) from a type 1 early infantile epileptic encephalopathy (EIEE1) patient with a hemizygous R330L mutation in the ARX gene. Stem Cell Res. 60, 102693 (2022). [DOI] [PubMed]
38.Carss, K. et al. Comprehensive Rare Variant Analysis via Whole-Genome Sequencing to Determine the Molecular Pathology of Inherited Retinal Disease. Am. J. Hum. Genet.100, 75–90 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Swain, P. K. et al. Mutations in the Cone-Rod Homeobox Gene Are Associated with the Cone-Rod Dystrophy Photoreceptor Degeneration. Neuron19, 1329–1336 (1997). [DOI] [PubMed] [Google Scholar]
40.Li, L. et al. Detection of variants in 15 genes in 87 unrelated Chinese patients with Leber congenital amaurosis. PLoS One6, e19458 (2011). [DOI] [PMC free article] [PubMed]
41.Zhou, A. et al. Paired-like homeobox gene (PHOX2B) nonpolyalanine repeat expansion mutations (NPARMs): genotype–phenotype correlation in congenital central hypoventilation syndrome (CCHS). Genet. Med.23, 1656–1663 (2021). [DOI] [PubMed] [Google Scholar]
42.Parodi, S. et al. Parental origin and somatic mosaicism of PHOX2B mutations in Congenital Central Hypoventilation Syndrome. Hum. Mutat.29, 206 (2008). [DOI] [PubMed] [Google Scholar]
43.Webb, J. A. et al. Cooperative Gsx2–DNA binding requires DNA bending and a novel Gsx2 homeodomain interface. Nucleic Acids Res52, 7987–8002 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Torrado, M. et al. Role of conserved salt bridges in homeodomain stability and DNA binding. J. Biol. Chem.284, 23765–23779 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Clarke, N. D. Covariation of residues in the homeodomain sequence family. Protein Sci.4, 2269–2278 (1995). [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Cho, G., Nasrallah, M. P., Lim, Y. & Golden, J. A. Distinct DNA binding and transcriptional repression characteristics related to different ARX mutations. Neurogenetics13, 23–29 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Barrera, L. A. et al. Survey of variation in human transcription factors reveals prevalent DNA binding changes. Science (1979)351, 1450–1454 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Shoubridge, C., Tan, M. H., Seiboth, G. & Gécz, J. ARX homeodomain mutations abolish DNA binding and lead to a loss of transcriptional repression. Hum. Mol. Genet.21, 1639–1647 (2012). [DOI] [PubMed] [Google Scholar]
49.Lei, T. et al. Prenatal exome sequencing in fetuses with callosal anomalies. Prenat. Diagn.42, 744–752 (2022). [DOI] [PubMed] [Google Scholar]
50.Uyanik, G. et al. ARX Mutations in X-Linked Lissencephaly with Abnormal Genitalia. Neurology61, 232–235 (2003). [DOI] [PubMed]
51.Kitamura, K. et al. Mutation of ARX causes abnormal development of forebrain and testes in mice and X-linked lissencephaly with abnormal genitalia in humans. Nat. Genet.32, 359–369 (2002). [DOI] [PubMed] [Google Scholar]
52.Kato, M. et al. Mutations of ARX Are Associated with Striking Pleiotropy and Consistent Genotype-Phenotype Correlation. Human Mutation23 147–159 (2004). [DOI] [PubMed]
53.Papuc, S. M. et al. The role of recessive inheritance in early-onset epileptic encephalopathies: a combined whole-exome sequencing and copy number study. Eur. J. Hum. Genet.27, 408–421 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Hughes, A. E. O., Myers, C. A. & Corbo, J. C. A massively parallel reporter assay reveals context-dependent activity of homeodomain binding sites in vivo. Genome Res28, 1520–1531 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Zheng, Y., Sun, C., Zhang, X., Ruzycki, P. A. & Chen, S. Missense mutations in CRX homeodomain cause dominant retinopathies through two distinct mechanisms. 12, RP87147 (2023). [DOI] [PMC free article] [PubMed]
56.Weirauch, M. T. et al. Determination and inference of eukaryotic transcription factor sequence specificity. Cell158, 1431–1443 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Bateman, A. et al. UniProt: the Universal Protein Knowledgebase in 2025. Nucleic Acids Res53, D609–D617 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Bodenhofer, U., Bonatesta, E., Horejš-Kainrath, C. & Hochreiter, S. Msa: An R package for multiple sequence alignment. Bioinformatics31, 3997–3999 (2015). [DOI] [PubMed] [Google Scholar]
59.Pelé, J., Bécu, J. M., Abdi, H. & Chabbert, M. Bios2mds: An R package for comparing orthologous protein families by metric multidimensional scaling. BMC Bioinform.13, 133 (2012). [DOI] [PMC free article] [PubMed]
60.Wagih, O. Ggseqlogo: A versatile R package for drawing sequence logos. Bioinformatics33, 3645–3647 (2017). [DOI] [PubMed] [Google Scholar]
61.Cain, B., Decoding the mechanisms of cooperative DNA binding by the Paired-like homeodomain family, PairedLikeCooperativity, 10.5281/zenodo.20127595, 2026 [DOI] [PubMed]
62.Srivastava, D. et al. Molecular basis of CRX/DNA recognition and stoichiometry at the Ret4 response element. Structure (2024) 10.1016/j.str.2024.07.004. [DOI] [PMC free article] [PubMed]
63.Joint Center for Structural Genomics (JCSG) & Partnership for Stem Cell Biology (STEMCELL). Crystal structure of a distal-less homeobox protein 5 (Dlx5) from Homo sapiens at 1.85 A resolution. Worldwide Protein Data Bank 10.2210/pdb4rdu/pdb (2014). [DOI]
64.Hovde, S., Abate-Shen, C. & Geiger, J. H. Crystal Structure of the Msx-1 Homeodomain/DNA Complex. Biochemistry40, 12013–12021 (2001). [DOI] [PubMed] [Google Scholar]
65.Morgunova, E. et al. Interfacial water confers transcription factors with dinucleotide specificity. Nat. Struct. Mol. Biol.32, 650–661 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Eisenberg, D., Schwarz, E., Komaromy, M. & Wall, R. Analysis of Membrane and Surface Protein Sequences with the Hydrophobic Moment Plot. J. Mol. Biol. 179, 125–142 (1984). [DOI] [PubMed]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

All data generated in this study are found in the article and source data file. Source data are provided with this paper.

All custom scripts can be found at GitHub[https://github.com/cainbn97/PairedLikeCooperativity]⁶¹.

[CR1] 1.Lambert, S. A. et al. The Human Transcription Factors. Cell172, 650–665 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR2] 2.Jolma, A. et al. DNA-binding specificities of human transcription factors. Cell152, 327–339 (2013). [DOI] [PubMed] [Google Scholar]

[CR3] 3.Berger, M. F. et al. Variation in Homeodomain DNA Binding Revealed by High-Resolution Analysis of Sequence Preferences. Cell133, 1266–1276 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR4] 4.Badis, G. et al. Diversity and complexity in DNA recognition by transcription factors. Science (1979)324, 1720–1723 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR5] 5.Kock, K. H. et al. DNA binding analysis of rare variants in homeodomains reveals homeodomain specificity-determining residues. Nat. Commun. 15, 3110 (2024). [DOI] [PMC free article] [PubMed]

[CR6] 6.Bürglin, T. R. & Affolter, M. Homeodomain proteins: an update. Chromosoma125, 497–521 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR7] 7.Jun, S. & Desplan, C. Cooperative interactions between paired domain and homeodomain. Development122, 2639–2650 (1996). [DOI] [PubMed] [Google Scholar]

[CR8] 8.Slattery, M. et al. Cofactor binding evokes latent differences in DNA binding specificity between hox proteins. Cell147, 1270–1282 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR9] 9.Cain, B. et al. Prediction of cooperative homeodomain DNA binding sites from high-throughput-SELEX data. Nucleic Acids Res51, 6055–6072 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR10] 10.Qu, S., Tucker, S. C., Zhao, Q. & Wisdom, R. Physical and genetic interactions between Alx4 and Cart1. Development369, 359–369 (1999). [DOI] [PubMed] [Google Scholar]

[CR11] 11.Cai, R. L. Human CART1, a paired-class homeodomain protein, activates transcription through palindromic binding sites. Biochem. Biophys. Res. Commun.250, 305–311 (1998). [DOI] [PubMed] [Google Scholar]

[CR12] 12.Wilson, D., Sheng, G., Lecuit, T., Dostatni, N. & Desplan, C. Cooperative dimerization of paired class homeo domains on DNA. Genes Dev.7, 2120–2134 (1993). [DOI] [PubMed] [Google Scholar]

[CR13] 13.Tucker, S. C. & Wisdom, R. Site-specific heterodimerization by paired class homeodomain proteins mediates selective transcriptional responses. J. Biol. Chem.274, 32325–32332 (1999). [DOI] [PubMed] [Google Scholar]

[CR14] 14.Mazzoni, E. O. et al. Synergistic binding of transcription factors to cell-specific enhancers programs motor neuron identity. Nat. Neurosci.16, 1219–1227 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR15] 15.Durbin, A. D. et al. Selective gene dependencies in MYCN-amplified neuroblastoma include the core transcriptional regulatory circuitry. Nat. Genet.50, 1240–1246 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR16] 16.Boeva, V. et al. Heterogeneity of neuroblastoma cell identity defined by transcriptional circuitries. Nat. Genet.49, 1408–1413 (2017). [DOI] [PubMed] [Google Scholar]

[CR17] 17.Kim, S. et al. DNA-guided transcription factor cooperativity shapes face and limb mesenchyme. Cell187, 692–711 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR18] 18.Cain, B., Yuan, Z., Thoman, E., Kovall, R. A. & Gebelein, B. The ALX4 dimer structure provides insight into how disease alleles impact function. Nat. Commun.16, 4800 (2025). [DOI] [PMC free article] [PubMed]

[CR19] 19.Zheng, Y., Stormo, G. D. & Chen, S. Aberrant homeodomain–DNA cooperative dimerization underlies distinct developmental defects in two dominant CRX retinopathy models. Genome Res. 10.1101/gr.279340.124 (2024) [DOI] [PMC free article] [PubMed]

[CR20] 20.Treisman, J., Gonczy, P., Vashishtha, M., Harris, E. & Desplan, C. A Single Amino Acid Can Determine the DNA Binding Specificity of Homeodomain Proteins. Cell59, 553–562 (1989). [DOI] [PubMed]

[CR21] 21.Stenson, P. D. et al. The Human Gene Mutation Database (HGMD®): optimizing its use in a clinical diagnostic or research setting. Hum. Genet.139, 1197–1207 (2020). [DOI] [PMC free article] [PubMed]

[CR22] 22.Stenson, P. D. et al. The Human Gene Mutation Database: towards a comprehensive repository of inherited mutation data for medical research, genetic diagnosis and next-generation sequencing studies. Hum. Genet.136, 665–677 (2017). [DOI] [PMC free article] [PubMed]

[CR23] 23.Wilson, D. S., Guenther, B., Desplan, C. & Kuriyan, J. High resolution crystal structure of a paired (Pax) class cooperative homeodomain dimer on DNA. Cell82, 709–719 (1995). [DOI] [PubMed] [Google Scholar]

[CR24] 24.Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature630, 493–500 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR25] 25.Rohs, R. et al. The role of DNA shape in protein-DNA recognition. Nature461, 1248–1253 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR26] 26.Tsai, C. J., Lin, S. L., Wolfson, H. J. & Nussinov, R. Studies of protein-protein interfaces: A statistical analysis of the hydrophobic effect. Protein Sci.6, 53–64 (1997). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR27] 27.Trochet, D. et al. PHOX2B Genotype Allows for Prediction of Tumor Risk in Congenital Central Hypoventilation Syndrome. Am. J. Hum. Genet.76, 421–426 (2005). [DOI] [PMC free article] [PubMed]

[CR28] 28.Tucker-Kellogg, L. et al. Engrailed (Gln50→Lys) Homeodomain-DNA Complex at 1.9 Å Resolution: Structural Basis for Enhanced Affinity and Altered Specificity. Structure5, 1047–1054 (1997). [DOI] [PubMed]

[CR29] 29.Kissinger, C. R., Liu, B., Martin-Blanco, E., Kornberg, T. B. & Pabo, C. O. Crystal structure of an engrailed homeodomain-DNA complex at 2.8 Å resolution: A framework for understanding homeodomain-DNA interactions. Cell63, 579–590 (1990). [DOI] [PubMed] [Google Scholar]

[CR30] 30.Wilson, D. S., Sheng, G., Jun, S. & Desplan, C. Conservation and Diversification in Homeodomain-DNA Interactions: A Comparative Genetic Analysis. Proc. Natl. Acad. Sci. USA93, 6886–6891 (1996). [DOI] [PMC free article] [PubMed]

[CR31] 31.Holland, P. W. H., Booth, H. A. F. & Bruford, E. A. Classification and nomenclature of all human homeobox genes. BMC Biol. 5, (2007). [DOI] [PMC free article] [PubMed]

[CR32] 32.Thai, M. H. N. et al. Constraint and conservation of paired-type homeodomains predicts the clinical outcome of missense variants of uncertain significance. Hum. Mutat.41, 1407–1424 (2020). [DOI] [PubMed] [Google Scholar]

[CR33] 33.Trochet, D. et al. Germline Mutations of the Paired-Like Homeobox 2B (PHOX2B) Gene in Neuroblastoma. Am. J. Hum. Genet.74, 761–764 (2004). [DOI] [PMC free article] [PubMed]

[CR34] 34.Trochet, D. et al. Molecular consequences of PHOX2B missense, frameshift and alanine expansion mutations leading to autonomic dysfunction. Hum. Mol. Genet.14, 3697–3708 (2005). [DOI] [PubMed] [Google Scholar]

[CR35] 35.Strømme, P. et al. Mutations in the human ortholog of Aristaless cause X-linked mental retardation and epilepsy. Nat. Genet.30, 441–445 (2002). [DOI] [PubMed] [Google Scholar]

[CR36] 36.Monies, D. et al. The landscape of genetic diseases in Saudi Arabia based on the first 1000 diagnostic panels and exomes. Hum. Genet.136, 921–939 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR37] 37.Gong, X. et al. Generation of an isogenic gene-corrected iPSC line (OGHFUi001-A-1) from a type 1 early infantile epileptic encephalopathy (EIEE1) patient with a hemizygous R330L mutation in the ARX gene. Stem Cell Res. 60, 102693 (2022). [DOI] [PubMed]

[CR38] 38.Carss, K. et al. Comprehensive Rare Variant Analysis via Whole-Genome Sequencing to Determine the Molecular Pathology of Inherited Retinal Disease. Am. J. Hum. Genet.100, 75–90 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR39] 39.Swain, P. K. et al. Mutations in the Cone-Rod Homeobox Gene Are Associated with the Cone-Rod Dystrophy Photoreceptor Degeneration. Neuron19, 1329–1336 (1997). [DOI] [PubMed] [Google Scholar]

[CR40] 40.Li, L. et al. Detection of variants in 15 genes in 87 unrelated Chinese patients with Leber congenital amaurosis. PLoS One6, e19458 (2011). [DOI] [PMC free article] [PubMed]

[CR41] 41.Zhou, A. et al. Paired-like homeobox gene (PHOX2B) nonpolyalanine repeat expansion mutations (NPARMs): genotype–phenotype correlation in congenital central hypoventilation syndrome (CCHS). Genet. Med.23, 1656–1663 (2021). [DOI] [PubMed] [Google Scholar]

[CR42] 42.Parodi, S. et al. Parental origin and somatic mosaicism of PHOX2B mutations in Congenital Central Hypoventilation Syndrome. Hum. Mutat.29, 206 (2008). [DOI] [PubMed] [Google Scholar]

[CR43] 43.Webb, J. A. et al. Cooperative Gsx2–DNA binding requires DNA bending and a novel Gsx2 homeodomain interface. Nucleic Acids Res52, 7987–8002 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR44] 44.Torrado, M. et al. Role of conserved salt bridges in homeodomain stability and DNA binding. J. Biol. Chem.284, 23765–23779 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR45] 45.Clarke, N. D. Covariation of residues in the homeodomain sequence family. Protein Sci.4, 2269–2278 (1995). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR46] 46.Cho, G., Nasrallah, M. P., Lim, Y. & Golden, J. A. Distinct DNA binding and transcriptional repression characteristics related to different ARX mutations. Neurogenetics13, 23–29 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR47] 47.Barrera, L. A. et al. Survey of variation in human transcription factors reveals prevalent DNA binding changes. Science (1979)351, 1450–1454 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR48] 48.Shoubridge, C., Tan, M. H., Seiboth, G. & Gécz, J. ARX homeodomain mutations abolish DNA binding and lead to a loss of transcriptional repression. Hum. Mol. Genet.21, 1639–1647 (2012). [DOI] [PubMed] [Google Scholar]

[CR49] 49.Lei, T. et al. Prenatal exome sequencing in fetuses with callosal anomalies. Prenat. Diagn.42, 744–752 (2022). [DOI] [PubMed] [Google Scholar]

[CR50] 50.Uyanik, G. et al. ARX Mutations in X-Linked Lissencephaly with Abnormal Genitalia. Neurology61, 232–235 (2003). [DOI] [PubMed]

[CR51] 51.Kitamura, K. et al. Mutation of ARX causes abnormal development of forebrain and testes in mice and X-linked lissencephaly with abnormal genitalia in humans. Nat. Genet.32, 359–369 (2002). [DOI] [PubMed] [Google Scholar]

[CR52] 52.Kato, M. et al. Mutations of ARX Are Associated with Striking Pleiotropy and Consistent Genotype-Phenotype Correlation. Human Mutation23 147–159 (2004). [DOI] [PubMed]

[CR53] 53.Papuc, S. M. et al. The role of recessive inheritance in early-onset epileptic encephalopathies: a combined whole-exome sequencing and copy number study. Eur. J. Hum. Genet.27, 408–421 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR54] 54.Hughes, A. E. O., Myers, C. A. & Corbo, J. C. A massively parallel reporter assay reveals context-dependent activity of homeodomain binding sites in vivo. Genome Res28, 1520–1531 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR55] 55.Zheng, Y., Sun, C., Zhang, X., Ruzycki, P. A. & Chen, S. Missense mutations in CRX homeodomain cause dominant retinopathies through two distinct mechanisms. 12, RP87147 (2023). [DOI] [PMC free article] [PubMed]

[CR56] 56.Weirauch, M. T. et al. Determination and inference of eukaryotic transcription factor sequence specificity. Cell158, 1431–1443 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR57] 57.Bateman, A. et al. UniProt: the Universal Protein Knowledgebase in 2025. Nucleic Acids Res53, D609–D617 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR58] 58.Bodenhofer, U., Bonatesta, E., Horejš-Kainrath, C. & Hochreiter, S. Msa: An R package for multiple sequence alignment. Bioinformatics31, 3997–3999 (2015). [DOI] [PubMed] [Google Scholar]

[CR59] 59.Pelé, J., Bécu, J. M., Abdi, H. & Chabbert, M. Bios2mds: An R package for comparing orthologous protein families by metric multidimensional scaling. BMC Bioinform.13, 133 (2012). [DOI] [PMC free article] [PubMed]

[CR60] 60.Wagih, O. Ggseqlogo: A versatile R package for drawing sequence logos. Bioinformatics33, 3645–3647 (2017). [DOI] [PubMed] [Google Scholar]

[CR61] 61.Cain, B., Decoding the mechanisms of cooperative DNA binding by the Paired-like homeodomain family, PairedLikeCooperativity, 10.5281/zenodo.20127595, 2026 [DOI] [PubMed]

[CR62] 62.Srivastava, D. et al. Molecular basis of CRX/DNA recognition and stoichiometry at the Ret4 response element. Structure (2024) 10.1016/j.str.2024.07.004. [DOI] [PMC free article] [PubMed]

[CR63] 63.Joint Center for Structural Genomics (JCSG) & Partnership for Stem Cell Biology (STEMCELL). Crystal structure of a distal-less homeobox protein 5 (Dlx5) from Homo sapiens at 1.85 A resolution. Worldwide Protein Data Bank 10.2210/pdb4rdu/pdb (2014). [DOI]

[CR64] 64.Hovde, S., Abate-Shen, C. & Geiger, J. H. Crystal Structure of the Msx-1 Homeodomain/DNA Complex. Biochemistry40, 12013–12021 (2001). [DOI] [PubMed] [Google Scholar]

[CR65] 65.Morgunova, E. et al. Interfacial water confers transcription factors with dinucleotide specificity. Nat. Struct. Mol. Biol.32, 650–661 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR66] 66.Eisenberg, D., Schwarz, E., Komaromy, M. & Wall, R. Analysis of Membrane and Surface Protein Sequences with the Hydrophobic Moment Plot. J. Mol. Biol. 179, 125–142 (1984). [DOI] [PubMed]

PERMALINK

Decoding the mechanisms of cooperative DNA binding by the Paired-like homeodomain family

Brittany Cain

Connor Wasmund

Fiona C Rowan

Brian Gebelein

Abstract

Introduction

Results

Broad variation in cooperativity levels across the Paired-like class

Fig. 1. Paired-like factors vary in their ability to cooperatively bind the P3 site.

Residue differences across the Paired-like class contributes to cooperativity variation

Fig. 2. Paired-like HD sequence analysis, structural modeling, and quantitative DNA binding assays identify five HD residues that contribute to low cooperativity.

Table 1.

HD position 50 impacts both DNA binding cooperativity and specificity

Fig. 3. Dramatic differences in cooperativity between Q50 and K50 HD proteins.

Sequence constraints between ANTP and Paired-Like HDs identifies additional residues critical for cooperative DNA binding

Fig. 4. Paired-like and ANTP classes have different sequence constraints in the turn between helix 1 and helix 2 and within select residues of the recognition helix.

The role of HD positions 44 and 46 in cooperativity

The relationship between HD helical spacing and P3 site cooperativity

Fig. 5. The P26L Paired-like to ANTP class swap found in the ARX disease variant decreases P3 site cooperativity.

Defining rules of HD cooperativity on P3 sites

Paired-like factor pathogenic variants impact P3 site cooperativity

Fig. 6. Pathogenic variants in Paired-like factors impact cooperative binding.

Discussion

Sequence constraint identifies HD residues that contribute to cooperative DNA binding and specificity

Elucidating the impact of Paired-like HD cooperativity on protein function and disease variant mechanisms

Methods

Protein preparation

Electrophoretic mobility shift assays

AlphaFold3 predictions

Sequence constraint analysis

Helical spacing analysis

Statistics and reproducibility

Reporting summary

Supplementary information

Source data

Acknowledgements

Author contributions

Peer review

Peer review information

Funding

Data availability

Code availability

Competing interests

Footnotes

Contributor Information

Supplementary information

References

Associated Data

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases