Disparate binding kinetics by an intrinsically disordered domain enables temporal regulation of transcriptional complex formation

Significance

Different combinations and permutations of transcription factors work together to regulate the expression of target genes. These proteins often contain high levels of intrinsically disordered regions, which are important mediators of protein–protein interactions. We show that unusual binding kinetics associated with an intrinsically disordered region in a transcriptional coregulator can regulate the formation of transcriptional complexes that lead to the specification of neuronal cell subtypes. Notably, a single intrinsically disordered region shows selective differences in binding kinetics for proteins of the same family, which have implications for how intrinsic disorder contributes to regulatory processes and complexity in higher organisms.

Abstract

Intrinsically disordered regions are highly represented among mammalian transcription factors, where they often contribute to the formation of multiprotein complexes that regulate gene expression. An example of this occurs with LIM-homeodomain (LIM-HD) proteins in the developing spinal cord. The LIM-HD protein LHX3 and the LIM-HD cofactor LDB1 form a binary complex that gives rise to interneurons, whereas in adjacent cell populations, LHX3 and LDB1 form a rearranged ternary complex with the LIM-HD protein ISL1, resulting in motor neurons. The protein–protein interactions within these complexes are mediated by ordered LIM domains in the LIM-HD proteins and intrinsically disordered LIM interaction domains (LIDs) in LDB1 and ISL1; however, little is known about how the strength or rates of binding contribute to complex assemblies. We have measured the interactions of LIM:LID complexes using FRET-based protein–protein interaction studies and EMSAs and used these data to model population distributions of complexes. The protein–protein interactions within the ternary complexes are much weaker than those in the binary complex, yet surprisingly slow LDB1:ISL1 dissociation kinetics and a substantial increase in DNA binding affinity promote formation of the ternary complex over the binary complex in motor neurons. We have used mutational and protein engineering approaches to show that allostery and modular binding by tandem LIM domains contribute to the LDB1_LID binding kinetics. The data indicate that a single intrinsically disordered region can achieve highly disparate binding kinetics, which may provide a mechanism to regulate the timing of transcriptional complex assembly.

Intrinsically disordered regions (IDRs) are protein domains that lack a well-defined 3D structure. IDRs are frequently involved in making protein–protein interactions. It has been suggested that interactions involving IDRs offer many advantages over those involving solely structured domains, including the combination of high specificity with low affinity (K_d), increased sensitivity to environmental conditions and posttranslational modifications, increased flexibility, and the ability to provide a hub for multiple interactions (1, 2). Studies of IDR binding kinetics suggest that disorder can increase both association and dissociation rate constants (k_on and k_off, respectively) (3–6). Eukaryotic transcription factors are highly enriched in IDRs compared with other eukaryotic proteins and prokaryotic transcription factors (7). Despite their biological importance, relatively few studies report quantitative data for IDR-mediated transcription factor interactions (8).

The LIM-homeodomain (LIM-HD) and LIM-only (LMO) proteins provide a model for the role of IDRs in transcriptional complexes. All LIM-HD/LMO proteins contain two LIM domains (LIM1+2) arrayed in tandem that take part in protein–protein interactions. LIM-HD proteins also contain a central DNA binding homeodomain (HD) and an intrinsically disordered C-terminal domain (Fig. 1A). LIM domain binding protein 1 (LDB1) interacts with all LIM-HD/LMO proteins through a disordered LIM interaction domain (LID) (Fig. 1A), which folds on binding to LIM1+2 domains to form extended modular complexes (9–11). Competition for LDB1 by LIM-HD/LMO proteins contributes to a so-called transcriptional “LIM code” that helps determine cell fate in the developing spinal cord (12). A binary complex comprising LDB1 bound to the LIM-HD protein LHX3 triggers differentiation of V2-interneurons (V2-INs) (Fig. 1B) (13). In neighboring cells, a ternary complex is formed comprising LDB1, LHX3, and a second LIM-HD protein, ISL1. Here, LDB1_LID contacts ISL1_LIM1+2, forcing LHX3_LIM1+2 to bind a LID in the C-terminal domain of ISL1. The ternary complex triggers differentiation of spinal motor neurons (sMNs) (Fig. 1C) (13, 14). Paralogues ISL2 and LHX4 can form similar complexes in developing sMNs (Fig. 1 B and C) (15–17). Despite low sequence identity (Fig. 1D), LDB1_LID and ISL1/2_LID form very similar structures when in complex with their partners (11). sMN progenitors also express LMO4, which has been shown to inhibit the formation of the binary complex over the ternary complex by competing with LHX3/4 for LDB1 (18, 19).

Fig. 1.

Protein–protein and protein–DNA interactions in the LIM code. (A) Domain structures of the LIM-HD/LMO proteins and LDB1: tandem LIM domains (LIM1+2), HD, LID, and self-association (SA) domain. The LID (broken box) in LIM-HDs has only been found in ISL1/2. (B) LDB1 and LHX3/4 form a binary complex to regulate V2-IN development. (C) LDB1 binds ISL1/2, which in turn, binds LHX3/4 to form a ternary complex that regulates sMN development. (D) Structure-based sequence alignment of the LDB1 and ISL1/2 LIDs.

A lack of quantitative protein–protein and protein–DNA binding data means that it is unclear what mechanism governs the regulation of ternary and binary complexes in sMNs. Efforts to quantify LID:LIM1+2 interactions were hampered, as the LIM1+2 domains aggregate when expressed in isolation. However, engineered “tethered complexes,” comprising the LIM domains fused to LIDs via a flexible linker, are soluble and stable (10, 20). We recently developed a FRET-based solution method to study these interactions (21). Monomeric FRET protein pairs mYPet and mECFP are fused to the termini of the interacting domains within a tethered complex that has a human rhinovirus 3C protease (HRV 3C)-cleavable peptide linker. After proteolysis of the linker, the loss of FRET by competition with a nonfluorescent LID peptide or by dilution of the complex is monitored (Fig. 2A), allowing the determination of a broad range of values for K_d, k_off, and inferred k_on.

Fig. 2.

Highly variable binding affinities and kinetics occur in the protein–protein and protein–DNA interactions of the LIM code. (A) Protein fusion constructs and assay design for the FRET-based interaction studies. (B–D) Representative LID:LIM1+2 FRET-based competition, dilution, and dissociation assays, respectively (n = 3–7). Coloring is consistent across these panels. (E) Relationship of experimentally determined k_off and inferred k_on to K_d for the LID:LIM1+2 interactions. (F and G) Representative EMSAs of the LHX3_HD and 2HDLL constructs binding to DNAb and DNAt, respectively. Construct schematics and coloring are consistent with Fig. 1C. Gels were imaged using fluorescein-labeled oligonucleotides. (H) EMSA binding curves for LHX3_HD and 2HDLL constructs binding to DNAb/i/t. Curves are representative of replicate experiments (n = 2–3).

Here, we report quantification of the protein–protein interactions involved in the sMN LIM code using solution FRET-based methods. The binding kinetics of the LDB1:ISL1/2 interactions are much slower than those of the other LID:LIM1+2 interactions in the system. We combined these data with measurements of protein–DNA interactions using EMSAs and modeled the changes in populations of complexes. The formation of sMN ternary complexes is likely to be reliant on both the unusual kinetics of LDB1 binding by the different LIM-HD proteins and the higher affinity of the ternary complex for its target DNA sites. Binding data for mutants and single LIM domains suggest that small differences in related ordered LIM domains can modulate the LDB1_LID binding mechanism to produce highly disparate binding kinetics for a single IDR.

Results

LID:LIM1+2 Interactions Have Disparate Affinities and Kinetics.

We measured K_d and k_off values for complex formation between the LIDs from LDB1, ISL1, and ISL2 and the LIM1+2 domains from ISL1, ISL2, LHX3, and LHX4 (Table 1). The homologous competition approach, in which MBP-LDB1_LID was titrated into cleaved fluorescent complexes, was used to estimate the stronger binding affinities (K_d ≤ 2 nM) (Fig. 2B), and dilution experiments were used to estimate weaker binding affinities (K_d > 10 nM) (Fig. 2C). k_off values were determined by titrating with a large excess of MBP-LDB1_LID (Fig. 2D). k_off values for slowly dissociating complexes (k_off < 10⁻⁵ s⁻¹) were fitted using a fixed final FRET efficiency taken from the faster (k_off ∼ 10⁻⁴ s⁻¹) experiments.

Table 1.

Binding parameters for protein–protein interactions in the LIM code

LID	LIM1+2	Kd (M)	koff (s−1)	kon (M−1 s−1)
LDB1	ISL1	2.9 ± 0.3 × 10−8	2.86 ± 0.02 × 10−5	1.0 ± 0.1 × 103
LDB1	ISL2	2.5 ± 0.1 × 10−8	1.88 ± 0.02 × 10−5	7.6 ± 0.3 × 102
LDB1	LHX3	4.1 ± 0.4 × 10−10	6.3 ± 0.1 × 10−4	1.5 ± 0.2 × 106
LDB1	LHX4	1.0 ± 0.1 × 10−9	1.25 ± 0.01 × 10−3	1.3 ± 0.1 × 106
LDB1	LMO4	1.8 ± 0.1 × 10−9*	6.3 ± 0.02 × 10−4*	3.5 ± 0.2 × 105*
ISL1	ISL1	1.7 ± 0.3 × 10−4	N.D.	N.D.
ISL1	LHX3	2.7 ± 0.3 × 10−7	2.9 ± 0.3 × 10−1	1.1 ± 0.2 × 106
ISL1	LHX4	5.4 ± 0.5 × 10−8	6.8 ± 0.1 × 10−2	1.3 ± 0.1 × 106
ISL2	ISL2	4.4 ± 0.7 × 10−5	N.D.	N.D.
ISL2	LHX3	1.6 ± 0.3 × 10−7	2.2 ± 0.1 × 10−1	1.4 ± 0.4 × 106
ISL2	LHX4	3.3 ± 0.5 × 10−8	5.8 ± 0.7 × 10−2	1.8 ± 0.3 × 106

LID:LIM1+2 K_d and k_off values and inferred k_on were determined using FRET (n = 3–7). Constants represent the mean ± SEM from replicate experiments. ISL1/2 LID:LIM1+2 binding kinetics were not determined (N.D.).

^*Data for LDB1_LID:LMO4_LIM1+2 were reported in ref. 25.

The binding affinities of this network of like interactions span six orders of magnitude from the weakest intramolecular LID:LIM1+2 interactions in the ISL1/2 proteins (K_d ∼ 10⁻⁴–10⁻⁵ M) through the intermediate LDB1_LID:ISL1/2_LIM1+2 and ISL1/2_LID:LHX3/4_LIM1+2 (K_d ∼ 10⁻⁷–10⁻⁸ M) to the strongest LDB1_LID:LHX3/4_LIM1+2 interactions (K_d ≤ 10⁻⁹ M), which were similar to that previously reported for LDB1_LID:LMO4_LIM1+2 (21). The measured k_off values and inferred k_on values varied by up to four orders of magnitude (Table 1 and Fig. S1 A–C). It is generally expected that differences in K_d in like systems will be heavily influenced by differences in k_off. This holds true for the measured interactions of LHX3/4_LIM1+2 and LMO4 with LDB1_LID and ISL1/2_LID, as a linear trend line fits well to K_d and k_off for those interactions (Fig. 2E). However, the values for ISL1/2_LIM1+2 and LMO2 binding to LDB1_LID cluster together, well away from that trend line, due to large differences in k_on.

Binary and Ternary Complex Mimetics Have High DNA Binding Affinity and Specificity.

We used EMSAs to assess binding of HD-containing protein constructs to DNA, as more quantitative methods could not clearly distinguish between higher-affinity (presumably specific) and low-affinity (probably nonspecific) binding events (22). Nontagged isolated HDs from ISL1 and LHX3 were used to assess binding by individual proteins, whereas GST-tagged proteins and engineered fusion constructs provided models for dimeric binding. The oligonucleotide sequences used contain binding sites for ISL1, LHX3, and the ternary complex (henceforth referred to as DNAi, DNAb and DNAt, respectively).

A surprising outcome was that the ISL1_HD bound poorly to all oligonucleotides tested (Fig. S2 A–C). K_d estimates derived from the disappearance of the free oligonucleotide indicate that binding is in the range of K_d values approximately micromolar or higher (Tables 1 and 2 and Fig. S1D), which is at the limit of detection by EMSA. Use of a GST-ISL1_HD fusion had no effect on DNA binding affinity (Table 2). These results are consistent with genome occupancy studies, showing that in vivo DNA binding by ISL1 is dictated by its partner proteins (23, 24). In contrast, the isolated LHX3_HD bound strongly to DNAb but poorly to the other oligonucleotides (Fig. 2 F and H and Table 2). GST-LHX3_HD showed improved binding for DNAt but reduced binding to DNAb (Tables 1 and 2 and Fig. S2 D–F), highlighting the potential for binding artifacts in experimental systems (25). Incubation of both ISL1_HD and LHX3_HD with DNAt did not improve binding over the single proteins, suggesting that the isolated HDs do not bind cooperatively (Fig. S2G).

Table 2.

Binding parameters for protein–DNA interactions in the LIM code

HD	DNAi	DNAb	DNAt
ISL1HD	5.9 ± 0.3 × 10−6	2.0 ± 0.9 × 10−6	3 ± 2 × 10−6
GST-ISL1HD	6 ± 5 × 10−6	4 ± 4 × 10−6	>2 × 10−6*
LHX3HD	2 ± 1 × 10−6	9 ± 3 × 10−9	4 ± 3 × 10−6
GST-LHX3HD	>2 × 10−6*	5 ± 3 × 10−6	5 ± 5 × 10−7
2HDN	5 ± 2 × 10−7	4.4 ± 0.9 × 10−7	1.6 ± 0.4 × 10−7
2HDLL	7 ± 5 × 10−7	4 ± 2 × 10−6	2.4 ± 0.3 × 10−8

K_d values were determined using EMSA (n = 2–3). Constants represent the mean ± SEM from replicate experiments.

^*Data were more qualitative due to poor resolution of bands in some repeats (Fig. S2).

Fusion constructs mimicked the ternary complex: 2HDLL comprised ISL1_HD-LID and LHX3_LIM1+2-HD, whereas 2HDN comprised ISL1_HD and LHX3_HD. 2HDLL showed strong binding to DNAt and some evidence of weaker, probably nonspecific binding to DNAi and DNAb (Fig. 2 G and H, Table 2, and Fig. S2 H and I). The 2HDN fusion protein behaves similarly, albeit with slightly higher K_d for DNAt (Tables 1 and 2 and Fig. S2 J–L).

Formation of Binary Complex over Ternary Complex Is Favored at Equilibrium.

Competing protein–protein and protein–DNA binding events were simulated based on their equilibrium binding constants (Fig. S3). We used species concentrations of 1 nM to 100 µM to capture the common range of in vivo transcription factor concentrations (SI Materials and Methods), as nuclear concentrations of these proteins are not known. For each simulation, the total concentrations for all components were identical. LDB1 self-association and intramolecular ISL1/2 LID:LIM1+2 interactions were excluded from modeling.

The equilibrium states of LDB1, ISL1, and LHX3 were modeled in the absence of DNA. Here, the binary LDB1:LHX3 complex was the most populated species at all concentrations (Fig. 3A). Adding DNA showed varied effects. For V2-IN simulations (no ISL1/2), the binary complex bound its specific DNAb site with few off-target interactions (Fig. 3B and Fig. S4A). For sMN simulations, the presence of DNAt increased ternary complex formation relative to simulations without DNA, but the inclusion of DNAb led to a decrease in the ternary complex and an increase in the binary complex (Fig. 3B and Fig. S4 A and B). LMO4 is known to protect the ternary complex during sMN development by preferentially disrupting the binary complex (13, 18, 19). The addition of LMO4 to the equilibrium modeling resulted in the preferential formation of an LDB1:LMO4 complex, similar decreases in the ternary and binary complexes, and an increase in the transcriptionally unproductive ISL1:LHX3 complex (Fig. 3A and Fig. S4 B and C).

Fig. 3.

Equilibrium and kinetic modeling of the LIM code. (A) Complex formation in the sMN system at equilibrium modeled across various protein concentrations. (B) Comparison of the ternary and binary complexes bound to their specific DNA targets in the presence of different DNA sequences in the V2-IN (LDB1, LHX3, and DNAb) and sMN (LDB1, ISL1, LHX3, and DNAt) systems. Points and error bars represent the complex concentration calculated using the K_d and the range calculated from K_d ± SEM. (C) Kinetic modeling of the binary and ternary complexes after the addition of LHX3 to a preequilibrated LDB1:ISL1 complex. LMO4 was added 30, 60, 120, 150, or 200 min after the addition of LHX3. Each species concentration was set at 1 µM. (D) Comparison of t_C values from free and complexed simulations across different protein concentrations. (E) Competition native gel where mYPet-LDB1_LID was incubated with cleaved LDB1_LID:LIM1+2 complexes for the indicated times. The gel was imaged using mYPet fluorescence.

Sequestration of LDB1 by ISL1 Provides a Basis for Ternary Complex Formation.

We hypothesized that the unusual binding kinetics that we detected would resolve the apparent contradictions between in vivo data and our equilibrium modeling and investigated the effects of sequestration of LDB1 by ISL1. In these “free” simulations, LDB1 and ISL1 were equilibrated before the addition of LHX3, leading to rapid formation of the ternary complex followed by slower dissociation of LDB1:ISL1 and formation of the binary complex (Fig. 3C and Fig. S5 B and C). At 1 µM protein concentrations, the binary complex took 125 min to overtake the concentration of the ternary complex (henceforth referred to as the cross-over time; t_C) (Fig. 3C). As Lmo4 is up-regulated by the ternary complex (26, 27), we performed simulations in which LMO4 was added at intervals after the introduction of LHX3. Addition of LMO4 30–120 min after LHX3 caused a more than twofold increase in the t_C (Fig. 3C and Fig. S5D). That is, addition of LMO4 temporarily recreated the preferential disruption of the binary complex in both this kinetic model and in vivo.

To further investigate LDB1 sequestration in this system, additional simulations were performed assuming that LDB1 and ISL1 were fully complexed before addition of LHX3. These “complexed” simulations showed substantially increased t_C values at low protein concentrations (≤1 μM) but minimal effect at higher concentrations (≥10 μM) (Fig. 3D). Addition of LMO4 again increased t_C values (Fig. 3D). Equilibrium modeling using unequal protein concentrations showed that the highest ternary:binary ratio occurred with excess ISL1 and that the lowest occurred with excess LDB1 (Fig. S4D). Therefore, sequestration of LDB1 by ISL1 would allow a temporally stable ternary complex to form on addition or expression of LHX3.

To partially corroborate these results, we performed a competition native gel shift assay in which cleaved nonfluorescent complexes (LHX3_LIM1+2:LDB1_LID, GB1-ISL1_LIM1+2:LDB1_LID, and MBP-LMO4_LIM1+2:LDB1_LID) in various combinations were incubated with mYPet-LDB1_LID for either 5 or 100 min before separation on a native gel. The difference in molecular masses of the tags (GB1 ∼ 11 kDa, MBP ∼ 42 kDa) separates complexes that contain ISL1_LIM1+2, LHX3_LIM1+2, or LMO4_LIM1+2. Here, persistent binding of ISL1_LIM1+2 to LDB1_LID was evident when mYPet-LDB1_LID was incubated with GB1-ISL1_LIM1+2 for long periods before the addition of LHX3_LIM1+2 and MBP-LMO4_LIM1+2 (Fig. 3E), which is consistent with sequestration of LDB1 by ISL1 temporarily preventing the formation of other LDB1 complexes.

Allostery and Modularity of Tandem LIM Domains Contribute to LDB1_LID Binding.

We probed the mechanisms of disparate LID:LIM1+2 binding kinetics through several approaches. FRET-based binding experiments for LDB1_LID:ISL1_LIM1+2 and LDB1_LID:LHX3_LIM1+2 at varying NaCl concentrations tested the role of electrostatic steering. If this phenomenon was a major contributor to association, increased ionic strength would lead to a convergence in LDB1_LID k_on values for ISL1_LIM1+2 and LHX3_LIM1+2 (28). No such convergence was observed (Fig. S6 A–F and Table S2).

The sequence and structure of the LID:LIM1+2 binding interface and the overall folds of the complexes are very similar between the different LIM1+2 domains (Fig. S6 G, H, and K). A comparison of LIM1+2 sequences revealed five sites that correlated with binding kinetics, although these residues did not directly contact LDB1_LID (Fig. 4A and Fig. S6K). We made chimeric LIM1+2 domains that swapped the residues between the slow-binding LMO2 and the fast-binding LMO4. The chimeras had WT-like LDB1_LID K_d values, but their k_off values converged, such that the inferred LDB1_LID k_on of the chimeric LMO4_LIM1+2 was ∼7-fold higher than that of the chimeric LMO2_LIM1+2 compared with ∼170-fold for the WT domains (Fig. 4B, Fig. S6 I and J, and Table S3). These data indicated that the targeted sequence differences are important, but additional factors must contribute to disparate kinetics.

Fig. 4.

Investigation of the sequence and structural basis of disparate LDB1_LID binding properties by different LIM1+2 domains. (A) Residues in the LIM1+2 domains (gray) found to correlate with the differential LDB1_LID (cyan) binding are highlighted in red on the crystal structures of LDB1_LID:LMO2_LIM1+2 (Upper; Protein Data Bank ID code 2XJY) and LDB1_LID:LMO4_LIM1+2 (Lower; Protein Data Bank ID code 1RUT). (B) Inferred LDB1_LID k_on values for WT and chimeric LMO2/4_LIM1+2. (C) Single LIM FRET dilution assays for LDB1_LID:LMO2/4_LIM1/2. (D) Comparison of ΔG° for the LDB1_LID:LMO2/4_LIM1/2/1+2 interactions.

We measured ¹⁵N relaxation data of high k_on LDB1_LID:LHX3_LIM1+2 and low k_on LDB1_LID:LMO2_LIM1+2 for which we have partial NMR assignments to identify specific regions that might contribute to differences in binding. Overall, the datasets are very similar (Fig. S7A), but we could not assign peaks for part of a β-strand in LMO2_LIM1 that contacts LDB1_LID, suggesting that conformational exchange may be present. This region is highly ordered in LMO4_LIM1 (9), and comparisons of X-ray crystal structures suggest that a loop in this region may be more flexible than the rest of LMO2_LIM1+2 and LMO4_LIM1+2 (Fig. S7 C and D). Given these apparent differences in the LIM1 domains from the fast and slow complexes, we investigated the contributions of the separate LIM domains from LMO2 and LMO4 for binding to LDB1_LID. LDB1_LID:LMO2_LIM1 had the lowest K_d of the individual LIM domains (K_d = 110 ± 10 nM) (Fig. 4C), with k_off = 0.174 ± 0.006 s⁻¹ and an inferred k_on = 1.6 ± 0.2 × 10⁶ M⁻¹ s⁻¹ (Fig. S7E and Table S4). The other single LIM domains all had lower yet similar affinities for LDB1_LID (K_d ∼ 5–9 µM) (Fig. 4C and Table S4). The k_off values for these interactions could not be measured, as the complexes dissociated within the ∼1 s of dead time of the assay, implying k_off ≥ 3 s⁻¹ (Fig. S7 E and F) and an inferred k_on ≥ 10⁵ M⁻¹ s⁻¹, close to the k_on of LMO2_LIM1 and LMO4_LIM1+2 (∼10⁶ M⁻¹ s⁻¹) and well above that of LMO2_LIM1+2 (10³ M⁻¹ s⁻¹) (Table S4). The equilibrium data indicate that these LID:LIM1+2 interactions are subadditive compared with the individual LIM1 and LIM2 domains, particularly for LMO2 [ΔΔG°_{([LIM1+LIM2]−LIM1+2)} = 24.6 ± 0.6 kJ mol⁻¹ compared with 8.7 ± 0.4 kJ mol⁻¹ for LMO4] (Fig. 4D).

Discussion

IDRs can bind their ordered partners with k_on values that range from 10² to 10⁷ M⁻¹ s⁻¹ (28). The variation in k_on is at least partially determined by the folding propensities of the domain. For example, two IDRs, the transactivation domain (TAD) of c-Myb and the phosphorylated KID (pKID) domain from CREB, bind to the same region of the ordered KIX domain from CREB binding protein (CBP) using “mixed-conformational selection and induced folding” and nearly pure “induced folding” mechanisms, respectively (29, 30). The lower k_on of CREB_pKID relative to c-Myb_TAD correlates with the lower helical propensity of the pKID sequence (30). In contrast, we have found that a single IDR, LDB1_LID, can bind to related ordered partners with inferred k_on values that vary by up to three orders of magnitude (Tables 1 and 2). This variation in k_on is not based on differences in electrostatic steering (Fig. S6) (21) but instead, points to mechanistic differences in LDB1_LID binding conferred by the ordered LIM domains.

There is some precedence for ordered proteins influencing the folding and binding of their IDR partners (28). For example, mutations of the ordered MCL-1 open its binding pocket for the disordered BH3 motif from p53 upregulated modulator of apoptosis (PUMA), increasing k_on by approximately twofold (31), whereas mutations in CBP_KIX changed the transition state of complex formation with c-Myb_TAD, decreasing k_on by less than or equal to twofold (32). The large range of binding kinetics that LDB1_LID possesses for related LIM1+2 domains is highly unusual.

Our binding data for single LIM domain interactions from LMO2 and LMO4 suggest that modular binding to the flexible LDB1_LID drives the disparate kinetics. For LMO2, affinities for LDB1_LID suggest that the first step of binding is by LIM1, with a significant energetic penalty incurred for the subsequent binding of LIM2. The inferred LDB1_LID k_on of LMO2_LIM1+2 is much lower than for LMO2_LIM1/2 (Table S4), indicating that the binding of the second module is a rate-limiting step for complex formation. These barriers to association are likely caused by restrictions in the conformational changes available to LDB1 and LMO2. For LMO4, the individual LIM domains interact similarly with LDB1_LID, and although an energetic penalty is still incurred in binding both domains, this is smaller than it is for LMO2.

The origin of the differences in LDB1_LID binding kinetics may arise from a loop in LIM1 that appears to be flexible in LMO2 (low k_on) but not in LMO4 or LHX3 (high k_on), but the sequences of the loops are very similar (Fig. S7E). The same loop seems well-ordered in the crystal structure of ISL1 (low k_on), but other loops in ISL1_LIM2 appear to be more flexible (Fig. S7F), suggesting that the loops surrounding the β-strands that contact LDB1 contribute to binding kinetics. Notably, sequences that correlate with the disparate kinetics and were tested in chimeras tend to cluster away from the binding site, indicating allosteric regulation of LID:LIM1+2 binding. LIM domains are reported to contain extensive hydrogen bonding networks and high flexibility (33), which could facilitate this type of allostery.

The ability of a single IDR to selectively bind ordered partners with different binding kinetics is likely to be important for cellular events. Our kinetic and equilibrium simulations provide a model for how this could occur in the sMN LIM code. In our model, sequestration of LDB1 by ISL1 into a long-lived complex followed by the introduction of LHX3 would lead to the formation of a temporally stable ternary complex (Fig. 5). This timing is consistent with embryonic and induced pluripotent stem cell models of sMN development that show expression of ISL1 several days before LHX3, with both proteins expressed for several days, which would provide time for the transitions predicted in the kinetic model (34, 35). However, additional investigation of the in vivo kinetics of complex formation is required, as the rates constants are likely to differ as a result of factors, including crowded cellular environments and posttranslational modification. Nevertheless, we expect the general trends identified here to occur in vivo given the large relative differences in binding parameters. Note that the kinetic model but not the equilibrium model accounts for favored ternary complex formation and its regulation by LMO4 (13, 19). Expression of LMO4 and the HD transcription factor homeobox 9 (HB9) is up-regulated by the ternary complex (26, 27). The similar LDB1 binding rates of LHX3 and LMO4 teamed with the low LDB1:ISL1 k_off mean that LMO4 is able to compete with LHX3 to temporarily inhibit the formation of the binary complex but not the ternary complex, leaving LHX3 free to form the ternary complex. At the same time, HB9 competes with the binary complex for binding to DNA (18, 26), thus mimicking the conditions of the equilibrium modeling in the absence of DNAb, which favors ternary–DNA complex formation (Fig. 3B). When slow dissociation of the LDB1:ISL1 leads to the eventual breakdown of the ternary complex, HB9 and LMO4 would continue to prevent formation of the binary complex.

Fig. 5.

Mechanism for the kinetic and thermodynamic regulation of ternary complex formation. (A) LDB1 is sequestered by ISL1, disrupting the low-affinity intramolecular LID:LIM1+2 interaction and allowing the orphaned ISL1_LID to interact with LHX3_LIM1+2. (B) The intact ternary complex interacts with target DNA, up-regulating the expression of regulator proteins LMO4 and HB9. (C) LMO4 and HB9 cooperate to disrupt binary complex formation by competing for protein and DNA binding partners, respectively. SA, self-association.

Mechanisms other than timing of expression may contribute to LDB1:ISL1 sequestration. sMN progenitors express the E3 ubiquitin ligase RLIM, which targets LDB1 for proteasomal degradation, as well as ssDNA binding protein 1 (SSBP1), which protects LDB1:LIM-HD complexes from degradation and is required for full ternary complex activity (36–38). RLIM and SSBP1 could work together in these cells to protect nascent LDB1:ISL1 complexes and remove free LDB1 to prevent binary complex formation.

The expressions of ISL2 and LHX4 are also induced by the ternary complex and can promote sMN development (16, 19, 39, 40). Our binding data show ISL2_LID:LHX4_LIM1+2 has an eightfold higher K_d than the ISL1-LHX3 equivalent, resulting in higher ratios of ternary:binary complexes (Fig. S8). Thus, ISL2 and LHX4 could be even more effective that the canonical ISL1/LHX3 pair for sMN production from inducible pluripotent stem cells.

In summary, the modular and competing interactions between the intrinsically disordered LIDs and LIM domains arrayed in tandem result in variable binding kinetics between a single IDR and a family of related binding proteins. This phenomenon provides a mechanistic basis for the temporal regulation of transcription factor assembly and the resulting expression of sets of genes, with profound impacts on the development and evolution of complex organisms.

Materials and Methods

Construct Cloning, Expression, and Purification.

All constructs were generated by PCR using genes for mouse proteins and cloned into vectors for expression with 6×His, 6×His GB1, GST, or MBP tags for purification. The proteins were expressed in Escherichia coli BL21(DE3) or Rosetta II cells and purified by batch affinity chromatography followed by cation exchange and/or size exclusion chromatography (SI Materials and Methods).

Fluorescence Spectroscopy and FRET-Based Interaction Methods.

FRET-based LID:LIM1+2 assays were described previously (21). Experimental details acquisition parameters and binding equations are described in SI Materials and Methods.

EMSA.

Fluorescence EMSAs have been described previously (41). Oligonucleotides were designed to mimic in vivo promoters that were shown to bind either ISL1_HD or LHX3_HD alone or both HDs. More detail and binding equations are in SI Materials and Methods.

Native Gel Competition Assay.

Cleaved LID:LIM1+2 complexes (1 μM) were incubated with mYPet-LDB1_LID (50 nM) at room temperature for 5 or 100 min as indicated. Samples were resolved on a 4–16% acrylamide NativePAGE Bis⋅Tris gel (Thermo Fisher Scientific) in 100 mM Bis⋅Tris⋅tricine buffer at 150 V for 105 min. mYPet-fluorescence was visualized using the Typhoon FLA 9500 scanner (GE Healthcare) using the default GFP filter set with excitation at 473 nm and emission recorded at 510 nm.

Modeling Equilibria.

Competing protein–DNA and protein–protein interactions were modeled by numerical methods using either equilibrium or binding kinetics data using the MATLAB software (MathWorks). Binding affinities and rate constants were used to simulate competitive binding over a range of species concentrations (0.001–100 μM), with all species in a single simulation kept to equal total concentrations. The mean affinities and rate constants were used to calculate the amount of complex formed. To account for the errors in the EMSA data simulations involving DNA, species were repeated with K_d = mean ± SEM. The populations of the relevant species are displayed as the fraction bound. Equilibrium equations were reduced in Mathematica (Wolfram) before their numeric solving in MATLAB. The interactions and equations used are detailed in SI Materials and Methods.