A Structural Basis for the Regulation of the LIM-Homeodomain Protein Islet 1 (Isl1) by Intra- and Intermolecular Interactions*

Background: A putative intramolecular interaction in the Islet 1 (Isl1) transcription factor inhibits DNA binding. Results: An intramolecular interaction between the LIM domains and LIM homeobox 3 (Lhx3)-binding domain in Isl1 was characterized. Conclusion: The intramolecular interaction within Isl1 is weak but specific. Significance: This interaction likely prevents unproductive binding in the absence of cofactor proteins. Islet 1 (Isl1) is a transcription factor of the LIM-homeodomain (LIM-HD) protein family and is essential for many developmental processes. LIM-HD proteins all contain two protein-interacting LIM domains, a DNA-binding homeodomain (HD), and a C-terminal region. In Isl1, the C-terminal region also contains the LIM homeobox 3 (Lhx3)-binding domain (LBD), which interacts with the LIM domains of Lhx3. The LIM domains of Isl1 have been implicated in inhibition of DNA binding potentially through an intramolecular interaction with or close to the HD. Here we investigate the LBD as a candidate intramolecular interaction domain. Competitive yeast-two hybrid experiments indicate that the LIM domains and LBD from Isl1 can interact with apparently low affinity, consistent with no detection of an intermolecular interaction in the same system. Nuclear magnetic resonance studies show that the interaction is specific, whereas substitution of the LBD with peptides of the same amino acid composition but different sequence is not specific. We solved the crystal structure of a similar but higher affinity complex between the LIM domains of Isl1 and the LIM interaction domain from the LIM-HD cofactor protein LIM domain-binding protein 1 (Ldb1) and used these coordinates to generate a homology model of the intramolecular interaction that indicates poorer complementarity for the weak intramolecular interaction. The intramolecular interaction in Isl1 may provide protection against aggregation, minimize unproductive DNA binding, and facilitate cofactor exchange within the cell.


Islet 1 (Isl1) is a transcription factor of the LIM-homeodomain (LIM-HD) protein family and is essential for many developmental processes. LIM-HD proteins all contain two proteininteracting LIM domains, a DNA-binding homeodomain (HD), and a C-terminal region. In Isl1, the C-terminal region also contains the LIM homeobox 3 (Lhx3)-binding domain (LBD), which interacts with the LIM domains of Lhx3. The LIM domains of Isl1 have been implicated in inhibition of DNA binding potentially through an intramolecular interaction with or close to the HD. Here we investigate the LBD as a candidate intramolecular interaction domain. Competitive yeast-two hybrid experiments
indicate that the LIM domains and LBD from Isl1 can interact with apparently low affinity, consistent with no detection of an intermolecular interaction in the same system. Nuclear magnetic resonance studies show that the interaction is specific, whereas substitution of the LBD with peptides of the same amino acid composition but different sequence is not specific. We solved the crystal structure of a similar but higher affinity complex between the LIM domains of Isl1 and the LIM interaction domain from the LIM-HD cofactor protein LIM domainbinding protein 1 (Ldb1) and used these coordinates to generate a homology model of the intramolecular interaction that indicates poorer complementarity for the weak intramolecular interaction. The intramolecular interaction in Isl1 may provide protection against aggregation, minimize unproductive DNA binding, and facilitate cofactor exchange within the cell.
LIM-homeodomain (LIM-HD) 5 proteins are transcription factors that are critical in the development of many cell types and tissues. Mammals carry 12 LIM-HD genes, which exist as six pairs of closely related paralogues in each species: Isl1/Isl2, Lhx1/Lhx5, Lhx2/Lhx9, Lhx3/Lhx4, Lhx6/Lhx8, and Lmx1a/ Lmx1b. LIM-HD proteins and their binding partners are expressed in overlapping but distinct patterns to specify cell types or tissues at various locations and stages of development (e.g. Refs. 1 and 2). These properties have led to the concept of the "LIM code" in which LIM-HD proteins act in a combinatorial manner to regulate tissue patterning at the level of transcription (3,4).
All LIM-HD proteins contain two tandemly arrayed LIM domains at or near the N terminus that mediate protein-protein interactions followed by a homeodomain that binds DNA and a C terminus that is predicted to be unstructured (e.g. see Fig. 1). The activity of LIM-HD proteins is dependent on the essential cofactor LIM domain-binding protein 1 (Ldb1; for a review, see Ref. 5). Ldb1 contains an N-terminal self-association domain (6) that forms trimers in vitro (7) and mediates long range chromatin interactions (8 -10). It also contains a ϳ30-residue LIM interaction domain (LID) that binds to the LIM domains of all LIM-HD and closely related LIM-only (LMO) proteins (6,(11)(12)(13)(14). Ldb1 LID binds as an extended peptide across both LIM domains of its binding partners (13,15,16). Isl1 and Isl2 each contain a LID-like sequence in their C-terminal regions, the Lhx3-binding domain (LBD), which binds the LIM domains Lhx3 and Lhx4 (15,17). These four proteins are co-expressed in and are necessary for the formation of motor neurons in the ventral neural cord (2, 18 -20). By binding the LIM domains of Lhx3 and Lhx4 (Lhx3/4 LIM1ϩ2 ), Isl1 LBD and Isl2 LBD play a critical role in enabling the formation of ternary Ldb1⅐Isl1/2⅐Lhx3/4 complexes, 6 which in turn upregulate the expression of motor neuron-specific genes such as Hb9 (see Fig. 1A and Refs. 15 and 21-23).
Ldb1 LID and Isl1/2 LBD bind the LIM domains of Lhx3/4 in a similar manner (15,17), opening up the possibility that Isl1 LBD could mediate interactions with LIM domains from other LIM-HD proteins. Isl1 is critical to the development of motor neurons, pituitary, pancreas, heart, and sensory neurons of the retina and inner ear (18, 24 -29). Understanding the roles of this protein in development and the potential ways in which it may be modulated requires a complete appreciation of the behavior of Isl1. This includes the possibility that Isl1 may form an intramolecular interaction through the LBD with its own LIM domains (see Fig. 1B).
The presence of an intramolecular interaction within Isl1 was originally suggested by Sánchez-García et al. (30) based on evidence that the LIM domains in Isl1 could disrupt the DNA binding activity of the protein in vitro. Isl1 showed higher affinity and increased specificity of DNA binding when the LIM domains were removed or denatured. These observations were believed to be the result of an intramolecular interaction between Isl1 LIM1ϩ2 and Isl1 HD . The LIM domains of Lhx3, Mec3 (a Caenorhabditis elegans Lhx1/5 paralogue), and Xlim1 (Xenopus Lhx1) have also been observed to negatively regulate DNA binding of their respective homeodomains (12,(31)(32)(33). Dawid et al. (34,35) speculated that the LIM domains of LIM-HD proteins may bind at or near the homeodomain to inhibit DNA binding with inhibition being relieved by interaction of the LIM domains with another binding partner.
This work extends our previous discovery of the Isl1 LBD and characterization of its ability to bind to LIM domains from Lhx3 and Lhx4 (15) through the investigation of binding specificity and modeling of an intramolecular interaction between Isl1 LIM1ϩ2 and Isl1 LBD . We present binding and NMR evidence to demonstrate a weak but specific intramolecular interaction between those domains in Isl1 that provides a possible mechanism by which the LIM domains of Isl1 can influence DNA binding. We have further determined the crystal structure of Isl1 LIM1ϩ2 bound to Ldb1 LID and use this structure to generate a homology model of an Isl1 LIM1ϩ2 ⅐Isl1 LBD complex.

EXPERIMENTAL PROCEDURES
Cloning and Protein Expression-Protein residue numbering refers to the following National Center for Biotechnology Information (NCBI) mouse protein entries: Isl1, NP_067434.3; Ldb1, NP_034827.1; and Lhx3, NP_001034742.1. For LIM1ϩ2 constructs in Fig. 2   , NP_034855.1. Genes were cloned from mouse brain tissue or generated as synthetic genes (GenScript USA Inc.). Isl1 LIM1ϩ2 refers to residues 11-138, Isl1 LBD refers to residues 262-301, Ldb1 LID refers to residues 300 -330, and Lhx3 LIM1ϩ2 refers to residues 28 -153. All other constructs are defined as they appear in the text. Constructs were generated by PCR and sequenced to confirm identity (Australian Genome Research Facility, Westmead Millennium Institute, Sydney, Australia). Proteins were expressed with a GST tag using pGEX-2T (GE Healthcare) in Escherichia coli BL21 (DE3) cells at 20°C for 16 -20 h. All proteins were purified by glutathione-Sepharose 4B resin (GE Healthcare) and eluted by proteolytic cleavage of the tag with thrombin (Sigma-Aldrich) or the addition of 20 mM reduced glutathione (Sigma-Aldrich). The proteins were additionally purified by size-exclusion chromatography using a Superdex 200 16/60 column (GE Healthcare) in 20 mM Tris (pH 8.5), 150 mM NaCl, and 1 mM dithiothreitol (DTT).
For yeast mating experiments, AH109 cells were transformed with pGBT9 constructs, Y187 cells were transformed with pGAD10 constructs, and cells were grown overnight at 30°C in liquid media lacking tryptophan or leucine, respectively. Cultures of AH109 and Y187 cells at A 600 nm Ϸ1.0 were spotted on top of each other in an array on rich media and incubated at 30°C for ϳ16 h. Spots containing mated diploid cells were replica-plated onto moderate stringency media and incubated at 30°C for ϳ48 h.
Multi-angle Laser Light Scattering-Purified proteins were subjected to size exclusion chromatography using a Superose 12 10/30 size exclusion column (GE Healthcare) in line with a Wyatt refractometer and MiniDawn multiangle laser light scattering detector (Wyatt Technology). The concentration of protein as determined by the change of refractive index and the intensity of scattered light were used to experimentally determine the molecular mass passing through the detector.
Thermal Denaturation-The thermal stability of proteins was determined using the ThermoFluor method (36,37). Melting experiments were performed in a 96-well plate format using a 7500 Fast Real-Time PCR System (Applied Biosystems) in tetramethylrhodamine filter mode at a heating rate of 45 s°C Ϫ1 . Protein was used at a concentration of ϳ0.33 mg ml Ϫ1 in 20 mM Tris (pH 8.5), 150 mM NaCl, and 1 mM DTT and spiked with SYPRO Orange (Invitrogen) at a final concentration of ϳ16؋ (based on the "5000؋ stock solution" as supplied by Invitrogen). Fluorescence readings were taken at 1°intervals from 25 to 95°C. Data were normalized for basal fluorescence and expressed as a percentage of maximum fluorescence with minimum fluorescence treated as zero.
Nuclear Magnetic Resonance (NMR)-Spectra were acquired at 298 K on a 600-or 800-MHz spectrometer (Bruker Avance III) equipped with a 5-mm TCI CryoProbe (Bruker) as indicated in the figure legends. 15 N transverse relaxation optimized spectroscopy-heteronuclear single quantum coherence (TROSY-HSQC) experiments were performed using the standard pulse sequence trosyetf3gpsi from the Bruker library. Spectra were processed with TopSpin (Bruker) and analyzed with SPARKY. 7 X-ray Structure Determination and Refinement-The crystallization, collection, and processing of data for Isl1 LIM1ϩ2 -Ldb1 LID were described previously (38). The AutoSol (39) applet in the PHENIX software compilation (40) was used to generate experimental phases. The program PHASER (41) was used to identify the zinc atom positions and to generate initial phases with the anomalous data to 3.1-Å resolution. The program RESOLVE (42) was used for density modification and solvent flattening. The model was built manually from the resultant map and subsequently revised using Coot (43).
For refinement of the model, the data were reprocessed and scaled using HKL-2000 (44) to include only the first 360 images of 0.5°oscillation each, which corresponds to 180°total oscillation and half of the total collected images. Refinement using the new data set was carried out using REFMAC (45) with isotropic temperature factors, TLS groups, and metal-ligand bond restraints between the zinc atoms and coordinating atoms (46). The MOLPROBITY server was used for structure validation, identifying steric clashes, and geometric problems (47). Surfaces and contacts were evaluated using PISA (48) and LIGPLOT (49). The coordinates have been deposited in the Protein Data Bank (accession code 4JCJ).
Homology Model Construction-A homology model of Isl1 LIM1ϩ2 ⅐Isl1 LBD was built from existing structures using Coot (43). The models of chain B from Isl1 LIM1ϩ2 -Ldb1 LID and chain B from Lhx3 LIM1ϩ2 -Isl1 LBD (Protein Data Bank code 2RGT) were aligned by the ␣ carbons of each LIM domain separately. The residues of Ldb1 LID were substituted for the equivalent residues in Isl1 LBD : Ldb1 300 -310 became Isl1 263-273 and Ldb1 315-328 became Isl1 275-288 . Side chains of the substituted amino acids were put in the equivalent rotamer found in Lhx3 LIM1ϩ2 -Isl1 LBD where possible or the highest probability rotamer that resulted in a sensible steric arrangement.

Potential Interactions of the Lhx3-binding Domain-The
LIM domains from LIM-HD proteins tend to aggregate in vitro, preventing their use in standard biophysical interactions assays (50 -52). However, these proteins are well suited to yeast twohybrid assays. Thus, we used yeast two-hybrid analysis to determine whether Isl1 LBD could interact with the LIM domains of LIM-HD proteins other than Lhx3 and Lhx4, but we could only see strong evidence of interactions (robust yeast growth on media lacking histidine and blue color formation in the presence of X-␣-Gal) for these already characterized interactions ( Fig. 2A). Note that growth of yeast cells co-transformed with Isl2 and Isl1 LBD was indistinguishable from that of the control for Isl2 (cells co-transformed with Isl2 and an empty vector) and therefore does not indicate an interaction between Isl1 and Isl2.
The yeast two-hybrid assay has a lower limit of detection of K A ϳ 10 5 M Ϫ1 for this type of interaction (15,53). So despite a lack of evidence for an intermolecular interaction between the LIM domains of Isl1 (Isl1 LIM1ϩ2 ) and Isl1 LBD ( Fig. 2A and Ref. 15), we hypothesized that Isl1 LBD might be able to bind Isl1 LIM1ϩ2 in an intramolecular fashion. That is, an intrinsically weak interaction between those domains would be enhanced through decreased entropy loss brought about by the domains being part of the same polypeptide chain. Before investigating the existence of this weak intramolecular interaction, we tested whether any intramolecular interaction could be physically accommodated in the native protein (in which the homeodomain lies between LIM domains and LBD; Fig. 1B). Thus, we generated a construct of Isl1 that encompassed the LIM domains, homeodomain, and LBD (Isl1 11-291 ) and a chimera in which the Isl1 LBD was substituted by Ldb1 LID (Isl1ϩLdb1; Fig.  2, B and C). This differs from "tethered complexes" (Fig. 2D) characterized in previous studies in which the LIM domains from LIM-HD/LMO proteins are tethered to Ldb1 LID or Isl1/ 2 LBD via a flexible linker (e.g. Refs. 13, 15, 17, 53, and 54). We used the yeast two-hybrid system to compare the binding of these constructs to known protein partners, Lhx3 LIM1ϩ2 (which can bind Isl1 LBD or Ldb1 LID ) and Ldb1 LID (which can bind Isl1 LIM1ϩ2 ) (15). Whereas Isl1 11-291 was able to bind robustly to each of Lhx3 LIM1ϩ2 and Ldb1 LID under all selection conditions, Isl1ϩLdb1 did not bind its targets under any conditions (Fig. 2E). This result is consistent with a strong intramolecular interaction between Isl1 LIM1ϩ2 and Ldb1 LID in the chimera that prevents intermolecular interaction with either Lhx3 LIM1ϩ2 or Ldb1 LID . That is, an LBD-like sequence can physically contact Isl1 LIM1ϩ2 despite the presence of the intervening homeodomain in the primary sequence, suggesting that an intramolecular LIM-LBD interaction could physically occur in the Isl1 protein. However, the data for Isl1  indicate that the competing intermolecular interactions Isl1 LBD ⅐Lhx3 LIM1ϩ2 and Isl1 LIM1ϩ2 ⅐Ldb1 LID would be much stronger than a putative intramolecular interaction. A, the ternary complex that specifies motor neurons is formed between the proteins Ldb1 (green), Isl1 (blue), and Lhx3 (cyan) (15,23). Ldb1 can self-associate to form trimers (7). LIM, LIM domain; SD, self-association domain. B, schematic of the proposed intramolecular LIM-LBD interaction in Isl1.

Generation of Tethered Complexes and Negative Control
LBDs-Stable forms of LIM-HD and LMO proteins are generated by tethering the LIM domains to an interaction partner such as Ldb1 LID or Isl1 LBD (15,51,52,55,56) (Fig. 2D). This approach works well because those peptide-like structures bind their target LIM domains in a head-to-tail manner (i.e. the C terminus of the LID/LBD lies near the N terminus of the LIM domains and vice versa). We generated Isl1 LIM1ϩ2 -peptide complexes to further investigate the intramolecular Isl1 and intermolecular Isl1⅐Ldb1 interactions. Thus, we engineered Isl1 LIM1ϩ2 -Isl1 LBD and Ldb1 LID -Isl1 LIM1ϩ2 constructs. We generated LIM1ϩ2-linker-peptide ("LIMs first") and peptidelinker-LIM1ϩ2 ("LIMs second") variants for both complexes. Changing the order of binding partners in this type of complex can affect the apparent stability (57), so throughout this study we compare only like with like for tethered complexes (e.g. LIMs first with LIMs first or LIMs second with LIMs second) as indicated in the figure legends. For simplicity, we henceforth refer to the tethered complexes of Isl1 LIM1ϩ2 with Ldb1 LID and Isl1 LBD as Isl1-Ldb1 LID and Isl1-Isl1 LBD , respectively, regardless of the order of binding partners.
To establish whether Isl1 LBD binds Isl1 LIM1ϩ2 in a specific or nonspecific manner, we generated two negative control LBDs based on Isl1 LBD . In "Switched," the positions of the LIM1-and LIM2-binding motifs within the LBD sequence were switched (Fig. 3A). As the two binding motifs from LID/LBD-like peptides are thought to not be interchangeable (58), this design should prevent Switched from contacting both LIM1 and LIM2 simultaneously and thereby reduce binding affinity. Note that for this construct an interaction could be mediated with LIM1 or LIM2. "Scrambled" was generated by randomizing the order of the 30 residues of Isl1 LBD . This design maintains the same overall physical properties but should disrupt any contacts that confer specific binding. Switched and Scrambled were tested for binding against Lhx3 LIM1ϩ2 in a yeast two-hybrid assay, and as expected, both showed reduced binding activity (Fig. 3B). Switched and Scrambled were incorporated into tethered Isl1 LIM1ϩ2 -LBD constructs as described above, resulting in Isl1-Switched and Isl1-Scrambled.
Blocking Ldb1 LID with Tethered Complexes-We used a competitive yeast two-hybrid approach to establish the presence of a weak intramolecular interaction in Isl1. The assay used Isl1-Isl1 LBD constructs (including wild-type, switched, and scrambled versions of LBD) fused to the Gal4 activation domain (AD) and Ldb1 LID peptides (including a series of Ldb1 LID peptides that have reduced affinity for Isl1 LIM1ϩ2 ; Ref. 15) fused to the Gal4 DNA-binding domain (DBD). Ldb1 LID peptides that can effectively compete with Isl1 LBD for binding to Isl1 LIM1ϩ2 will disrupt the intramolecular interaction and result in activation of reporter genes, whereas peptides that have a reduced affinity for Isl1 LIM1ϩ2 and cannot effectively compete with Isl1 LBD for binding to Isl1 LIM1ϩ2 will not activate reporter genes (Fig. 3C). Thus, in terms of assessing the relative strength of the intramo-lecular interaction, reduced levels of yeast growth indicate a stronger interaction, and robust levels of growth indicate a weaker interaction.
Wild-type Ldb1 LID was able to bind all three tethered constructs under both moderate and high stringency conditions, , and high (ϪHisϪAde) stringency. "ϩϩϩ" indicates robust growth at three yeast densities (A 600 nm ϭ 0.2, 0.02, and 0.002), "ϩϩ"indicates growth at the two higher densities, "ϩ"indicates growth at the highest density only, "ϳ"indicates minor levels of growth only at the highest density, and "Ϫ" indicates no growth at any dilution used. "ND" indicates growth in the negative control for that selection condition, so the binding strength could not be assessed under that condition. All tethered constructs tested are of the LIMs-second form.
indicating that neither Isl1 LBD nor the negative control LBDs can block the interaction of Ldb1 LID with Isl1 LIM1ϩ2 (Fig. 3D). Similarly, T323A did not significantly disrupt binding to any of the tethered complexes tested, reflecting its high affinity for Isl1 LIM1ϩ2 , as observed previously (15). Conversely, V303A/ I322A abolished binding to the tethered constructs under all conditions. All other mutants of Ldb1 LID bound Isl1-Isl1 LBD very weakly or not at all, indicating that an intramolecular Isl1⅐Isl1 LBD interaction exists that can out-compete weaker mutant Ldb1 LID ⅐Isl1 LIM1ϩ2 interactions. In contrast, the mutant peptides showed much higher levels of binding to Isl1-Scrambled and Isl1-Switched with the latter showing the highest levels of binding. These data indicate that the relative strengths of the binding affinities of Isl1 LBD for Isl1 LIM1ϩ2 are wild type Ͼ scrambled Ͼ switched. Production and Assessment of Isolated Tethered Complexes-Recombinant Isl1-Isl1 LBD and Isl1-Ldb1 LID were expressed in E. coli and purified albeit at much lower quantities for Isl1-Isl1 LBD (ϳ0.3 mg of protein/liter of culture) compared with Isl1-Ldb1 LID (3-4 mg; Ref. 59). In contrast, Isl1-Switched and Isl1-Scrambled either formed large aggregates or were degraded, precluding biophysical or structural analysis as isolated tethered complexes.
The Specificity of the Intramolecular Interaction-Fortunately, it was possible to express and purify modest amounts of 15 Nlabeled GST-tagged Isl1-Switched and Isl1-Scrambled. Thus, we recorded 15 N-1 H HSQC NMR spectra of the GST-tagged Isl1-peptide complexes to assess the folded states of these constructs. This approach has been used successfully in the past to assess the stability and folded state of GST-tagged proteins (60,61). In those studies, the linker between the GST and the protein of interest was flexible, and the proteins of interest had a low molecular mass (ϳ10 kDa or smaller). Traditional HSQC spectra are very poor at detecting signals from species with a high molecular mass due to slower tumbling of molecules in solution, so very few peaks were observed for the GST component (at ϳ50 kDa for dimeric GST) of those constructs, and the spectra predominantly reported narrow, sharp peaks for the faster tumbling small protein of interest. In comparison, the various tethered complexes containing Isl1 LIM1ϩ2 are larger (ϳ20 kDa) and expected to be rod-shaped. We collected 15 N-1 H TROSY-HSQC data (Fig. 5) because traditional HSQC spectra recorded on our GST fusion proteins displayed very few peaks (Ͻ5% of the number of peaks observed by the TROSY experiments). TROSY experiments exploit a partial cancellation of the dipole-dipole coupling of the amide group and chemical shift anisotropy of the nitrogen atoms for certain interactions (62). This increases signal resolution and allows the measurement of larger molecular mass proteins.
The spectra of the GST-tagged tethered complexes (GST-Isl1-Ldb1 LID , GST-Isl1-Isl1 LBD , GST-Isl1-Switched, and GST-Isl1-Scrambled) were compared with that of GST alone (Fig. 5). The 15 N-1 H TROSY-HSQC spectrum of GST yielded many more peaks than the standard 15 N-1 H HSQC (60), but these peaks are predominantly clustered in a few areas of the spectrum, and all four GST-tagged tethered complexes contained additional peaks when compared with GST alone. The spectrum of GST-Isl1-Ldb1 LID showed a reasonable number (ϳ75) of well dispersed and sharp non-GST peaks indicative of a highly folded protein (Fig. 5A). The spectrum of GST-Isl1-Isl1 LBD showed many of the same peaks as that of GST-Isl1-Ldb1 LID (ϳ50), but these peaks were generally broader (Fig.  5B). In contrast, the spectra for GST-Isl1-Switched and GST-Isl1-Scrambled contained far fewer peaks. Other than those for tryptophan side chains ( 15 N, ϳ130 ppm; 1 H, ϳ10 ppm), the only non-GST peaks that are visible for these proteins are poorly dispersed and lie in the "unfolded" regions of the spectra (Fig. 5, C and D) (63), suggesting that these constructs are not well folded and are probably molten globule-like. On the whole,  these data indicate that Ldb1 LID and Isl1 LBD can form specific interactions with the LIM domains of Isl1, but the negative control LBDs, Switched and Scrambled, do not.
The Crystal Structure of Isl1-Ldb1 LID -We attempted to determine the three-dimensional structure of Isl1-Ldb1 LID and Isl1-Isl1 LBD to better understand the nature of the intermolecular and intramolecular Isl1-peptide interactions. Isl1-Isl1 LBD did not yield diffraction quality crystals, but Isl1-Ldb1 LID was successfully crystallized, and diffraction data were collected as described previously (38).
Experimental phases were generated from this diffraction data by single wavelength anomalous dispersion at the zinc absorption edge (1.2186 Å) to a resolution of 3.1 Å (Table 1). To improve the quality of the data and reduce potential errors introduced by radiation damage, we reprocessed it to take the first 360 frames (instead of 720) and extended the resolution to 3.0 Å (Table 1). These data were used for refinement of the Isl1-Ldb1 LID model. The R work and R free of the final model correspond well with the resolution of the data and the amount of the electron density that can be adequately modeled (ϳ20% of total mass is missing; Table 1; Protein Data Bank code 4JCJ).
Three molecules of Isl1-Ldb1 LID exist in the asymmetric unit and exhibit the same basic structure. The LIM domains of Isl1 in this complex are very similar to those of other LIM-only and LIM-HD complexes (13,15,17,54). The two LIM domains each contain two zinc-coordinating modules (i.e. coordinated atoms Zn1-4), which comprise two ␤-hairpins followed by a short ␣-helix (Fig. 6A, blue ribbon). The LIM domains are separated by a short region known as the "hinge." Ldb1 LID binds both LIM domains in an extended, head-to-tail fashion but does so in a bipartite manner in which two peptide segments of eight and 10 residues contact the two separate LIM domains (green ribbon). Residues Met 302 -Leu 309 contact Isl1 LIM2 and residues Asp 318 -Asn 327 contact Isl1 LIM1 . The intervening residues, many of which are absent from the electron density, correspond to the "spacer," which if present would sit across from the hinge of Isl1. The LIM1-binding region forms a single extended ␤-strand that packs in an antiparallel manner against the second ␤-hairpins of the first and second zinc-coordinating modules (which contain Zn1 and Zn2, respectively). These ␤-sheet structures overlap such that Ldb1 318 -322 contacts Isl1 56 -60 and Ldb1 321-327 contacts Isl1 [27][28][29][30][31][32][33] . In contrast, the LIM2-binding region forms a ␤-strand-like structure with the second ␤-hairpin of the third zinc-coordinating module (Zn3) and a ␤-strand with the second ␤-hairpin of the fourth (Zn4). Ldb1 308 contacts Isl1 92 and Ldb1 302-304 contacts residues of Isl1 117-120 (green ribbon), resulting in a single, three-stranded ␤-sheet being created at the fourth zinc-coordinating module. The buried surface area of the interface between Isl1 LIM1ϩ2 and Ldb1 LID is ϳ1290 Å 2 .
The three molecules in the asymmetric unit (chains A, B, and C; chain B is shown in Fig. 6A) are largely elongated, exhibiting angles at the hinge/spacer of 169°for A, 175°for B, and 176°for C. Chains B and C are the most similar (root mean square deviation (r.m.s.d.), 0.61 Å), whereas chain A is more divergent from B and C, having r.m.s.d. values of 1.45 and 1.67 Å, respectively. When the individual LIM domains or the corresponding LIMbinding regions of the LBDs are aligned, the molecules are much more similar (Fig. 6, B and C Of the 173 residues in Isl1-Ldb1 LID , 23 are not found in any of the chains, whereas an additional five are only found in chains B and C, one is in chains A and C, and another one is only in chain C. The missing residues from Isl1-Ldb1 LID are largely from the linker and termini of the protein but are also residues of the spacer of Ldb1 LID not adjacent to the LIM-binding regions (311-315). Of those residues that were modeled, ϳ25% are missing side chains (for details of missing residues and atoms, see Protein Data Bank code 4JCJ).
One of the crystal packing contacts is mediated by a disulfide bond between Cys-56 residues from pairs of Isl1 molecules (Fig.  6D). All protein samples used for crystallization contained DTT; however, samples appeared to crystallize more readily with age. Crystals appeared in ϳ2 weeks for a fresh protein sample but in ϳ1 week for a sample that was ϳ1 month old. Thus, the disulfide bond may have facilitated crystallization, but it lies away from the Isl1 LIM1ϩ2 ⅐Ldb1 LID interface and is unlikely to affect the interaction. In chain A only, a contact between Isl1 Arg-94 and Ldb1 Val-303 was observed (Fig. 6E). As it was not observed in the other two chains, it was treated as a crystallization artifact.
Homology Model of the Intramolecular Interaction-The structure of Isl1-Ldb1 LID provided us with a good template for mod-eling the interface of the intramolecular interaction. Ldb1 LID and Isl1 LBD bind Lhx3 LIM1ϩ2 in almost exactly the same manner (15), so we systematically altered the residues of Ldb1 LID in the Isl1-Ldb1 LID structure to the corresponding binding residues from Isl1 LBD using the Lhx3-Isl1/Ldb1 structures as a guide. Based on this homology model, Isl1 LBD ⅐ Isl1 LIM1ϩ2 is predicted to have a surface area (ϳ1330 Å 2 ) similar to that of Ldb1 LID ⅐Isl1 LIM1ϩ2 (ϳ1290 Å 2 ). The model of Isl1-Isl1 LBD reveals no apparent impediments to binding (Fig. 6, F and G), but several differences suggest an overall poorer complementarity between Isl1 LIM1ϩ2 and Isl1 LBD , resulting in the weaker intramolecular interaction. Isl1 LBD ⅐Isl1 LIM1ϩ2 forms 19 interdomain hydrogen bonds compared with 26 for Ldb1 LID ⅐Isl1 LIM1ϩ2 , and the residues of Ldb1 LID that are buried in grooves between the zinc-binding modules of each LIM domain are smaller in Isl1 LBD (Ldb1 Val-304 corresponds to Isl1 Ala-267 , and Ldb1 Ile-322 corresponds to Isl1 Val-282 ).

DISCUSSION
The structure of Isl1-Ldb1 LID can be compared with existing structures of Lhx3-Ldb1 LID , Lmo2-Ldb1 LID , Lmo4-Ldb1 LID , Lhx3-Isl1 LBD , and Lhx4-Isl2 LBD complexes (13,15,17,54). The tethered complexes generally crystallize in different conformations as a result of flexibility in the hinge/spacer region (17,54,59), so it is only meaningful to compare the structures of individual LIM-LID/LBD "half-complexes," the folds of which are highly conserved (Table 2). In this context, the complexes are very similar, although the lengths of well defined ␤-strands in Ldb1 differ (Fig. 7A). In the LIM1 halves of the complexes, Ldb1 forms one long ␤-strand with Isl1, two shorter ␤-strands with Lmo2 and Lmo4, and a single short ␤-strand with Lhx3 (Fig.  7A). For the LIM2 halves of the complexes, Ldb1 LID forms a single short ␤-strand with Isl1 and Lhx3 and two short ␤-strands with Lmo2 and Lmo4 structures (Fig. 7A).
Isl1, Lmo2, and Lmo4 all bind Ldb1 preferentially through their LIM1 domains, whereas Lhx3 preferentially binds through its LIM2 domain (14 -17). A possible contribution for this difference can be seen around the key LIM1-binding residue, Ldb1 Ile-322 . Whereas LIM1 in Isl1, Lmo2, and Lmo4 provides the perfect binding surface for the mt rotamer of Ldb1 Ile-322 , the equivalent surface in Lhx3 LIM1 binds Ldb1 Ile-322 as the less common pt rotamer (Fig. 7, B and C). These differences appear to be derived from Lhx3 having a shorter Loop 7, which connects the first and second ␤-hairpins of the second zinc-coordinating module and lies adjacent Ldb1 Ile-322 (Fig.  7D). The longer Loop 7 also appears to enable the formation of the additional backbone-backbone hydrogen bonds across the Ldb1 LID ⅐Isl1/Lmo2/Lmo4 LIM1 interfaces that result in increased secondary structure in Ldb1 for Isl1/Lmo2/Lmo4 compared with Lhx3 at LIM1 (Fig. 7A).
Our data support the presence of an intramolecular interaction in Isl1 whereby the LBD from this protein can interact in a specific manner with its own LIM domains. Our yeast twohybrid and thermal denaturation data of the Isl1-Isl1 LBD tethered complex indicate that the intramolecular interaction is weak compared with intermolecular Isl1⅐Ldb1 binding, suggesting that the intermolecular interaction is favored. The origin of an intramolecular interaction between Isl1 LBD and Isl1 LIM1ϩ2 is unclear. It may simply reflect that Isl1 LBD binds the LIM domains of Lhx3 and Lhx4 on the same surface as Ldb1 LID , which in turn binds Isl1 LIM1ϩ2 . Alternatively, it is pos-sible that intramolecular interactions involving LIM domains were a feature of ancestral LIM-HD proteins in which case the Isl1 LBD may have subsequently evolved to mediate intermolecular interactions with Lhx3/4. Regardless, the intramolecular interaction could play one or more positive roles in regulating the function of Isl1, many of which benefit from having a weak rather than a strong binding affinity. Uncomplexed LIM domains from Isl1 are likely to be "sticky" because of exposed hydrophobic surfaces. Isl1 LBD might shield the LIM domains and prevent Isl1 LIM1ϩ2 from nonspecifically binding other proteins in the cell or prevent the protein from being targeted for proteolytic degradation. The intramolecular interaction may as suggested in concept by Dawid (35) prevent Isl1 from binding DNA in the absence of its FIGURE 6. The structure of Isl1-Ldb1 LID (Isl1 LIM1؉2 -Ldb1 LID ). A, ribbon structure of chain B of Isl1-Ldb1 LID with Isl1 in blue and Ldb1 in green. Zinc atoms (Zn1-4) are shown as gray spheres, and coordinating residues are shown as sticks. Missing residues of the engineered tether (linker) and middle of Ldb1 LID (spacer) are modeled as dashed lines. B and C, backbone alignment of the LIM domains for chains A (red), B (yellow), and C (blue) along LIM1 (B) and LIM2 (C). The construct in the crystal structure is of the LIMs-first form. D, a disulfide bond between Cys-56 from chains A and C links two molecules of Isl1-Ldb1 LID (ribbon). E, Arg-94 (chain A) forms hydrogen bonds (black dashed lines) with the backbone carbonyl oxygens from Val-303 (chain B) molecule and Glu-306 (chain A). Arg-94 (chain B) is not visible in the electron density. In D and E, Isl1 is shown in blue, and Ldb1 is shown in green in chains A and C with lighter colors for chain B. Cys-56, Arg-94, and residues 303-306 from Ldb1 are shown as Corey-Pauling-Koltun-colored sticks. A 2F o F c electron density map around the residues is shown as a mesh contoured at 1.5 . F and G, comparison of Isl1 LIM1ϩ2 ⅐Isl1 LBD and Isl1 LIM1ϩ2 ⅐Ldb1 LID interfaces at LIM1 (F) and LIM2 (G). Isl1 LBD (blue sticks with oxygen atoms in red and nitrogen atoms in dark blue) is modeled onto Isl1-Ldb1 LID in place of Ldb1 LID using an alignment of Lhx3-Isl1 LBD (Protein Data Bank code 2RGT) and Lhx3-Ldb1 LID (Protein Data Bank code 2JTN) as a guide. Ldb1 LID as modeled in the crystal structure is shown as transparent green sticks. Isl1 LIM1ϩ2 is shown as a white surface. other protein partners (Fig. 8A) but as a weak interaction allow readily for productive binding (Fig. 8B). We suggested previously that preferential binding of Isl1 LIM1 to Ldb1 and of Lhx3 LIM2 to Ldb1 and Isl1 might facilitate an exchange mechanism to explain how the Ldb1⅐Lhx3 complex could be disrupted by Isl1 (15). An intramolecular interaction between Isl1 LBD and Isl1 LIM1ϩ2 allows for protection of Isl1 LIM1ϩ2 as well as Lhx3 LIM1ϩ2 in this scenario with both the Isl1⅐Ldb1 and Isl1⅐Lhx3 interactions in the Ldb1⅐Isl1⅐Lhx3 ternary complex forming in parallel (Fig. 8C).
The possibility of an intramolecular interaction in Isl1 opens up the potential of similar interactions in other LIM-HD proteins especially if the origin of Isl1 LBD occurred in an ancestral LIM-HD. The largely uncharacterized and unstructured C-terminal regions of LIM-HD proteins stand in stark contrast to the highly structured and conserved N-terminal regions. It seems probable that in at least some LIM-HD proteins these regions contain a non-conserved, intrinsically unstructured sequence that might very weakly but specifically interact with the LIM domains in an intramolecular fashion. A, the intramolecular interaction between Isl1 LIM1ϩ2 and Isl1 LBD could impose steric effects on the homeodomain, impacting DNA binding. B, when Ldb1 LID binds Isl1, the intramolecular interaction is displaced (bottom), and this inhibition would be released. C, cofactor exchange mechanism. When Isl1 approaches the binary Ldb1⅐Lhx3 complex the intramolecular interaction between Isl1 LIM1ϩ2 and Isl1 LBD may facilitate the formation of an intermediate ternary complex in which the N-and C-terminal halves of Isl1 LBD and Ldb1 LID contact different molecules to enable Isl1 to efficiently displace Ldb1 LID ⅐Lhx3 LIM1ϩ2 and generate the ternary complex in a stepwise manner.