Implementing the LIM code: the structural basis for cell type-specific assembly of LIM-homeodomain complexes

Abstract

LIM-homeodomain (LIM-HD) transcription factors form a combinatorial ‘LIM code' that contributes to the specification of cell types. In the ventral spinal cord, the binary LIM homeobox protein 3 (Lhx3)/LIM domain-binding protein 1 (Ldb1) complex specifies the formation of V2 interneurons. The additional expression of islet-1 (Isl1) in adjacent cells instead specifies the formation of motor neurons through assembly of a ternary complex in which Isl1 contacts both Lhx3 and Ldb1, displacing Lhx3 as the binding partner of Ldb1. However, little is known about how this molecular switch occurs. Here, we have identified the 30-residue Lhx3-binding domain on Isl1 (Isl1_LBD). Although the LIM interaction domain of Ldb1 (Ldb1_LID) and Isl1_LBD share low levels of sequence homology, X-ray and NMR structures reveal that they bind Lhx3 in an identical manner, that is, Isl1_LBD mimics Ldb1_LID. These data provide a structural basis for the formation of cell type-specific protein–protein interactions in which unstructured linear motifs with diverse sequences compete to bind protein partners. The resulting alternate protein complexes can target different genes to regulate key biological events.

Introduction

The formation of cell type-specific complexes is crucial for the development of complex organisms. Unique combinations of LIM-homeodomain (LIM-HD) proteins are thought to form a transcriptional ‘LIM code' that is required for the specification of cell types within many different tissues and organs (reviewed in Gill, 2003). LIM-HD proteins are characterised by two tandemly arrayed LIM domains at or near their N termini that mediate interactions with other proteins (Bach, 2000), and a central homeodomain (HD) that recognises TAAT-containing DNA sequences. The LIM and HD regions share a high level of sequence conservation, but the C-terminal regions of the proteins are diverse.

The LIM domains from LIM-HD and related LIM-only (LMO) proteins bind to the LIM domain-binding (Ldb, also known as CLIM, NLI or CHIP) proteins, by means of an ∼30-residue LIM interaction domain (LID) on Ldb1 (Ldb1_LID; Jurata et al, 1998; Deane et al, 2004). This broadly expressed protein is involved in multiple developmental pathways (reviewed by Matthews and Visvader, 2003). In LMO:Ldb1 complexes, Ldb1_LID binds as an extended peptide, stretching across both LIM domains in a head-to-tail fashion (Deane et al, 2003, 2004). Ldb1 also contains an N-terminal self-association (SA) domain. Many of the biological activities of LIM-HD proteins depend on binding to Ldb1 oligomers, and at least two different LIM-HD or LMO proteins may simultaneously take part in Ldb1-containing complexes (Jurata et al, 1998).

Specific transcriptional codes involving LIM-HDs are particularly important in the developing central nervous system of vertebrates where these proteins are important in the specification of a large number of distinct cell types. For example, the two LIM-HD proteins, Lhx3 (LIM homeobox protein 3) and Isl1 (Islet-1), act together with Ldb1 to specify two distinct cell types that lie adjacent to each other in the developing spinal cord, namely V2 interneurons and motor neurons. Ldb1 and Lhx3 are both present in developing V2 interneurons, whereas Isl1 is additionally expressed in postmitotic motor neurons (Pfaff et al, 1996; Sharma et al, 1998). In the developing chick, the introduction of Isl1 into immature V2 interneurons gives rise to axonal outgrowths that are characteristic of motor neurons (Thaler et al, 2002). Thaler et al also provided compelling evidence that in V2 interneurons Lhx3 binds Ldb1 directly to form a transcriptionally active complex, whereas the additional presence of Isl1 in motor neurons results in a situation where Isl1 directly contacts Ldb1 and Lhx3.

Here we used a combination of structural, mutagenic and biophysical approaches to identify the region of Isl1 that binds the LIM domains of Lhx3 (Isl1_LBD) and to establish the structural basis for the specification of motor neurons by Lhx3, Isl1 and Ldb1. Our X-ray and NMR structures of Lhx3:Ldb1_LID and Lhx3:Isl1_LBD complexes demonstrate that, despite the low homology between Ldb1_LID and Isl1_LBD, the two proteins bind in an essentially identical bipartite manner. Simulations of complex formation for these proteins based on experimental binding data indicate that binary and ternary complexes are likely to target different genes.

Results

Identification of the Lhx3-binding domain on Isl1

The region of Ldb1 that binds LIM domains, Ldb1_LID, is well defined (Figure 1A; Jurata and Gill, 1997; Deane et al, 2004); however, the region on Isl1 that binds the LIM domains of Lhx3 was only roughly identified as lying C-terminal to the LIM domains in Isl1 (Figure 1B; Thaler et al, 2002). We therefore generated a series of deletion mutants of Isl1 and used yeast two-hybrid (Y2H) analysis to precisely delineate the Lhx3/Isl1 interaction. Our data revealed a 30-residue sequence in Isl1 that binds the tandem LIM domains of Lhx3. Isl1_262–291, hereafter termed the Lhx3-binding domain Isl1_LBD, lies C-terminal to the HD and N-terminal to the 16-residue Isl-specific domain (Thor and Thomas, 1997; Figure 1A and Supplementary data 1). Isl1_LBD and Ldb1_LID are essentially of the same length and share some slight sequence similarity near their N termini, but overall display very little sequence conservation. A far-UV CD spectrum of Isl1_LBD suggests that, similar to Ldb1_LID, the isolated domain is largely disordered (Figure 1C; Deane et al, 2004). Using Y2H, we showed that Isl1_LBD was not able to interact with tandem LIM domains of Lhx3 (Lhx3_LIM1+2) when the latter protein was expressed as a fusion with Ldb1_LID (Figure 1D; see below for a description of tethered complexes). Thus, we hypothesised that Ldb1 and Isl1 might contact the same or overlapping sites on Lhx3.

Figure 1.

Interactions of Isl1, Lhx3 and Ldb1 in developing V2 interneurons and motor neurons. (A) Schematics of Isl1, Lhx3 and Ldb1 from mice showing the arrangement of key domains. In addition to the LIM and HD domains, Isl1 contains the Lhx3-binding domain (LBD; residues 262–291; identified herein) and the Isl-specific domain (ISD, identified on the basis of sequence conservation 296–310; Thor and Thomas, 1997). (B) Relative arrangement of Ldb1 and Lhx3 in developing V2 interneurons (left) and Ldb1, Lhx3 and Isl1 in postmitotic motor neurons (right). The interactions examined in this paper are indicated with dashed circles (1, 2, 3). (C) Far-UV CD spectra for Isl1_LBD and Ldb1_LID at concentrations of ∼ 20 μM. Raw baseline-corrected data are presented. (D) Y2H data comparing the interaction of Isl1 (residues 133–349) with Lhx3_LIM1+2 in the absence or presence of Ldb1_LID. The selection conditions used were -L-W-H, 1 mM 3-AT.

Structure determination of Lhx3-binding domain complexes

To test the above hypothesis, we determined the structures of Lhx3/Ldb1 and Lhx3/Isl1 complexes. The isolated LIM domains from Lhx3, similar to most other LIM-HD and LMO proteins, tend to be insoluble and/or aggregation prone; however, we have developed a strategy to circumvent these problems by generating tethered proteins in which the LIM domains are fused to the Ldb1_LID by a flexible glycine/serine linker (Deane et al, 2001). Hence, we created tethered Lhx3_LIM1+2–Ldb1_LID and Lhx3_LIM1+2–Isl1_LBD complexes for structure determination. For the tethered Lhx3_LIM1+2–Ldb1_LID complex, we used restraints derived from multidimensional NMR data (Table I), whereas we used X-ray crystallography for the Lhx3_LIM1+2–Isl1_LBD complex (Table II; Supplementary data 2).

Table 1.

Statistics for the structural ensembles of Lhx3_LIM1+2– Ldb1_LID

Distance restraints
Total	2679
Intra-residue	1342
Sequential ((∣i−j∣)=1)	450
Medium range (∣i−j∣=2,3)	136
Long range (∣i−j∣>3)	742
Ambiguous	9
Dihedral angle restraintsa
ϕ	102
ψ	102
Atomic RMS differences (Å)	Backbone	Heavy atoms
Lhx328–89+Ldb1316–327b	0.70±0.11	1.18±0.12
Lhx391–151+Ldb1301–311	0.60±0.12	0.97±0.16
Lhx328–151	1.32±0.33	1.62±0.30
Ldb1301–327	1.25±0.32	1.60±0.30
Lhx328–151+Ldb1301–327	1.36±0.32	1.66±0.29
PROCHECK-NMR statistics
residues in
Favoured region	74.3%
Allowed region	23.3%
Generously allowed regions	1.5%
Disallowed regions	0.9%
Mean deviations from the ideal geometry
Bond length (Å)	0.00182±0.00147
Bond angles (deg)	0.2895±0.184
aFrom TALOS.
bThe structured regions of the complex are defined as Ldb1301–327 and Lhx328–151, which have order angle parameters for ϕ and ψ >0.9, with the exception of Ldb1312–315 and Lhx339–40,76–78,90,140−141.

Table 2.

Data collection and refinement statistics for Lhx3_LIM1+2–Isl1_LBD

	MAD data sets			Native data seta
	Peak	Remote	Inflection
Space group	C2
Unit cell parameters (Å, deg)	a=119, b=62.2, c=51.9, β=91.6
Wavelength (Å)	1.282	1.170	1.283	1.54
Resolution (Å)	2.30 (2.34–2.30)	2.30 (2.34–2.30)	2.30 (2.34–2.30)	2.05
Mosaicity (deg)	1.05	1.05	1.05	1.04
No. of unique reflections	14 610	14 313	14 339	23 109
Completeness (%)	85.4 (54.1)	83.5 (47.5)	83.7 (48.9)	96.5 (78.1)
Redundancy	6.4 (4.8)	6.1 (4.5)	6.3 (4.6)	3.5 (2.1)
Rmergeb	0.045 (0.218)	0.045 (0.243)	0.042 (0.215)	0.051 (0.388)
Average I/σ(I)	15.2 (7.0)	14.8 (6.2)	15.3 (7.3)	13.4 (2.4)
Phasing statistics
Resolution range (Å)	30–2.3
Zn sites/asymmetric unit	8
FOMMADc	0.66
FOMRESOLVEd	0.7
Model refinement
Rcryste	0.214 (0.287)
Rfreef	0.254 (0.346)
No. of reflections used in refinement	21 921
No. of reflections in the test set	1188 (5.1%)
Protein atoms (including Zn)	2377
Water molecules	48
RMSD bond length (Å)	0.02
RMSD bond angle (deg)	1.5
Mean protein B factor, all non-H atoms (Å2)	43.1
Mean water B factor (Å2)	43.5
Estimated standard uncertainties
Coordinates, based on residual R (Å)	0.2
Coordinates, based on Rfree (Å)	0.18
Ramachandran plot, residues in
Favoured regions (%)	93.9
Additional allowed regions (%)	6.1
Disallowed regions (%)	0
Values for the highest resolution shell are given in parentheses.
aNative set data from Bhati et al (2008).
bRmerge=∑h∑i∣Ii−〈I〉∣/∑h∑iIi.
cFigure of merit after SOLVE phasing.
dFigure of merit after RESOLVE.
eRcryst=∑∣∣Fobs∣−∣Fcalc∣∣/∑∣Fobs∣, where ∣Fobs∣ and ∣Fobs∣ are the observed and calculated structure factor amplitudes.
fRfree is Rcryst for the 5% validation set.

The solution structure of Lhx3_LIM1+2–Ldb1_LID

Lhx3_LIM1+2–Ldb1_LID forms an elongated complex in which Ldb1_LID binds along the length of the tandem LIM domains of Lhx3 and the engineered linker that tethers the two proteins is unstructured (Figure 2A). An overlay of the 20 lowest energy structures of Lhx3_LIM1+2–Ldb1_LID has a backbone RMSD of 1.4 Å over the structured regions of the complex (Lhx3_28–151 and Ldb1_301–327). However, each half of the complex is better defined, with backbone RMSDs of 0.70 Å for the ‘LIM1 half' (Lhx3_28–89 and Ldb1_316–327; Figure 2B) and 0.60 Å for the ‘LIM2 half' (Lhx3_91–151 and Ldb1_301–311; Figure 2C). Although ¹⁵N-¹H heteronuclear NOE data for the complex (Supplementary data 3A) do not reveal any region of increased flexibility between the two LIM domains, residues Ldb1_313–315 (which lie between the two LIM-binding segments) do have slightly increased mobility (¹⁵N-¹H NOE values of 0.4–0.6). Further, comparison with the structures of other double LIM domain complexes with LIM-binding peptides (i.e., the LMO4:Ldb1_LID complexes (Deane et al, 2004; Jeffries et al, 2006) and the Lhx3_LBD complex—see below) indicates that the two LIM domains can take up different relative orientations related by a hinge-like motion. These data suggest that the larger RMSD for the full complex might arise at least in part from a degree of mobility between the two LIM domains. In the ensemble, several residues that lie between the two halves of the complex, Lhx3₉₀ and Ldb1_312–315, are less well defined (order angle parameters for ϕ and ψ angles <0.8); these regions are hereafter referred to as the ‘hinge' in Lhx3 and ‘spacer' in Ldb1_LID.

Figure 2.

Structures of Lhx3_LIM1+2–LID complexes. (A) Ribbon representation of the lowest energy structure of Lhx3_LIM1+2–Ldb1_LID in solution. Only the structured regions of Lhx3 (blue) and Ldb1_LID (yellow) are shown. The zinc ions (grey spheres) and the side chains of the zinc-ligating residues (orange) are shown. The position of the unstructured linker is indicated. (B, C) Backbone traces of the 20 lowest energy structures of the Lhx3_LIM1+2–Ldb1_LID complex. Overlays over the backbone atoms of (B) LIM1 (Lhx3_28–89 plus Ldb1_316–327) and (C) LIM2 (Lhx3_91–151 and Ldb1_301–311) are shown. (D) The structured residues of monomeric Lhx3_LIM1+2–Isl1_LBD complex. Lhx3 is shown in blue and Isl1_LBD is shown in green. (E) The symmetry-related dimer Lhx3_LIM1+2–Isl1_LBD. Half-dimers are coloured green and blue, respectively, with Lhx3 in the darker and Isl1 in the lighter colour. (F) Overlay of dimeric (Lhx3 (green)/Isl1_LBD (turquoise)) and monomeric (Lhx3 (yellow)/Isl1_LBD (magenta)) forms of the Lhx3_LIM1+2–Ldb1_LID complex overlaid over the backbone atoms of Lhx3_LIM2. Only residues 262–273 of Isl1_LBD from the dimer are shown for clarity. Figures were prepared in MolMol (Koradi et al, 1996).

To confirm that the tethered Lhx3_LIM1+2–Ldb1_LID construct mimicked the native complex, we used a construct that contained a Factor Xa protease site in the linker. ¹⁵N-HSQC spectra from this variant before and after protease treatment are essentially identical (Supplementary data 3B and C), demonstrating that the conformations of Lhx3 and Ldb1_LID are equivalent in the intra- and intermolecular complexes.

The crystal structure of Lhx3_LIM1+2–Isl1_LBD

The structure of the Lhx3_LIM1+2–Isl1_LBD complex was determined using native and multiple anomalous dispersion data (collected at the Zn X-ray absorption edge) recorded to 2.05 and 2.30 Å resolution, respectively. The R and R_free values presented in Table II are consistent for an X-ray crystal structure to 2.05 Å resolution with disordered regions arising from the missing loop between Lhx3_LIM1+2 and the Isl1_LBD. There were two molecules of Lhx3–Isl1_LBD in the asymmetric unit, a monomer and a half-dimer, where the other half of the dimer comes from a symmetry-related molecule (Figure 2D and E). The structured regions of the monomer comprise Lhx3 residues 28–153 (plus two N-terminal residues, GS, derived from the vector) and Isl1 residues 262–288, forming an extended rod-like complex that resembles the solution structure of Lhx3_LIM1+2–Ldb1_LID (Figure 2A and D). The dimer comprises Lhx3 residues 32–154 and Isl1 residues 262–286, and is effectively a domain-swapped version of the monomer. The N-terminal half of Isl1_LBD (residues 262–273) binds its own intramolecular partner at the LIM2 domain of Lhx3, but is bent around by nearly 180° through Isl1_LBD residues 274–278 (the Isl1_LBD spacer) such that the remaining half of the Isl1_LBD segment (residues 279–286) contacts the LIM1 domain of Lhx3 in the symmetry-related molecule. Data from gel filtration monitored by multiangle laser light scatting revealed that the protein is monomeric in solution (Supplementary data 4), indicating that the 1:1 complex represents the dominant biological species (see Discussion for possible functional hints suggested by the domain-swapped dimer).

When the LIM1 domains of the different oligomeric forms of the Lhx3–Isl1_LBD complex are overlaid, it can be seen that the orientation of the two LIM domains with respect to each other differs; the LIM2 domain of the dimer has swung ‘downwards' from the hinge between the domains by ∼40° (Figure 2F). However, apart from differences that appear to be associated with the spacer residues in Isl1_LBD and the hinge region of Lhx3, the two halves of the different oligomeric forms are essentially identical: the backbone RMSDs for Lhx3_LIM1 (residues 32–89) are 0.86 Å and for Lhx3_LIM2 (residues 90–152) are 0.83 Å, and the contacts between Lhx3 and Isl1 are identical.

Geometry of the Lhx3_LIM1+2/peptide complexes

The LIM domains from the different structures conform to the typical LIM domain topology (Perez-Alvarado et al, 1996; Supplementary data 5), and are very similar to each other. RMSDs over the backbone residues of Lhx3 from the lowest energy member of the NMR ensemble of Lhx3–Ldb1_LID and the Lhx3–Isl1_LBD monomer are 1.8 Å for Lhx3_LIM1 and 1.3 Å for Lhx3_LIM2. However, the relative orientation of the two LIM domains differs in the different complexes (Supplementary data 5).

In the Lhx3–Ldb1_LID and Lhx3–Isl1_LBD structures, the Isl1/Ldb1 peptides bind both LIM domains from Lhx3 in an extended manner, forming β-strand(s) that pack in an antiparallel fashion against the second β-hairpin in each of the first and fourth Zn-ligating modules (Zn1 and Zn4, respectively; Figure 2A and D). Isl1_LBD forms an additional β-strand that packs against the equivalent hairpin in the second Zn-ligating module in Lhx3 (Zn2), and interdomain backbone–backbone hydrogen bonds suggest that some β-strand is also forming in Ldb1_LID where it packs against the second β-hairpin in the third Zn-ligating module (Zn3; Figure 3A). In the case of the Lhx3–Ldb1 complex, 3500–3600 Ų of surface area is buried at the interface between the two proteins, whereas the Lhx3–Isl1 complex buries ∼3250 Ų. Both interactions involve a combination of main-chain and side-chain hydrogen bonds, hydrophobic interactions and some electrostatic interactions (Figure 3A and B).

Figure 3.

Comparison of Lhx3-bound Ldb1_LID and Isl1_LBD. Interaction maps of (A) Lhx3_LIM1+2 and Ldb1_LID and (B) Lhx3_LIM1+2 and Isl1_LBD. Residues from Ldb1_LID are shown as yellow boxes and those from Isl1_LBD as green boxes. Residues from Lhx3 that form contacts with the LIDs are classed as indicated. (C, D) Structural alignment over the backbone atoms of (C) Lhx3_LIM1 and (D) Lhx3_LIM2 showing the side-chain heavy atoms of Isl1_LBD (green) and Ldb1_LID (orange). The Lhx3 LIM domains from the Isl1_LBD structure are shown as a grey surface. (E) Structure-based sequence alignment of Isl1_LBD and Ldb1_LID. Asterisks show residues that occupy equivalent positions in the structures. Key Lhx3-binding residues in (F) Ldb1_LID and (G) Isl1_LBD. Lhx3_LIM1+2 in each case is shown as a surface model (grey) with Ldb1_LID (yellow) or Isl1_LBD (green). For Ldb1_LID, the side chains of key residues from alanine scanning mutagenesis screens are classed as having a strong (red), moderate (orange) or weak (yellow) effect, whereas for Isl1_LBD the indicated residues (green) all have a strong effect.

Despite low sequence homology, the backbone and many of the side-chain atoms of the Lhx3-binding peptides occupy identical positions on the surface of Lhx3 (Figure 3C and D). The main differences occur within the spacers of the binding domains. A structure-based alignment of those domains (Figure 3E) reveals two binding sites of nine and seven residues, separated by a variable length (six and nine residues in Isl1_LBD and Ldb1_LID, respectively) spacer. The increased flexibility noted above for residues in the Ldb1_LID spacer is also consistent with the poor alignment of these residues between Ldb1_LID and Isl1_LBD and indicates that this stretch does not have an important role in recognition.

Mutagenic analysis of the interface

We used alanine scanning mutagenesis to look for key binding determinants of peptide/LIM-HD interactions. Sets of three residues in Ldb1_LID and Isl1_LBD were systematically mutated to alanine (and subsequently single and double point mutants were made) and tested for binding to LIM domain constructs of Lhx3 using Y2H (Table III and Figure 3F and G).

Table 3.

Mutagenic scanning of Ldb1_LID and Isl1_LBD

Mutations in either the N- or C-terminal halves of Ldb1_LID were able to reduce binding to Lhx3_LIM1+2. Ldb1_LID-V303, which packs against the surface of Lhx3_LIM2 and is only partially buried, had the strongest effect when mutated to alanine. I322 and M302, which are highly buried in the interface between Ldb1 and Lhx3_LIM1 and Lhx3_LIM2, respectively, had a more moderate effect when mutated. V304 and L309 were identified as having a weak effect on the interaction when mutated. These residues are buried in the Ldb1/Lhx3_LIM2 interface. Only Lhx3_LIM2 and not Lhx3_LIM1 was able to independently bind Ldb1_LID, indicating that although both ‘halves' of the interaction contribute to binding, the main binding determinants lie between the N-terminal half of Ldb1 and Lhx3_LIM2.

Only mutations in the N-terminal half of Isl1_LBD reduced binding to Lhx3_LIM1+2, and only Lhx3_LIM2 was able to independently bind the Isl1 peptide, showing that the main binding determinants between Isl1 and Lhx3 also lie in the N-terminal half of the peptide-binding domain and Lhx3_LIM2. The residues in Isl1_LBD that when mutated had the most effect on binding were M265 and A267, which are both highly buried at the interface. Notably, the key binding residues for the two complexes from our mutagenic data occupy very different physical spaces: Isl1_A267 is buried in a hydrophobic pocket formed between Zn1 and Zn2 in Lhx3_LIM1, whereas Ldb1_V303 lies flat on the surface of Lhx3_LIM1 (Figure 3F and G).

The peptides were also tested for binding against the LIM domains of Isl1. There was no evidence of an intermolecular interaction between Isl1_LIM1+2 and Isl1_LBD. However, in contrast to Lhx3, Isl1_LIM1 but not Isl1_LIM2, could mediate an interaction with Ldb1_LID. Mutation of several residues in the C-terminal half of Ldb1_LID impaired binding to Isl1_LIM1+2 and Isl1_LIM1, but no mutation in the N-terminal half of Ldb1_LID perturbed binding to Isl1_LIM1+2. These data indicate that the most important contacts are made between Isl1_LIM1 and the C-terminal half of Ldb1_LID, which is consistent with published glutathione-S-transferase (GST) pulldown data for this interaction (Jurata et al, 1996). However, mutations made against a background of the I322A mutation revealed that residues in the N-terminal half of Ldb1 (especially residues M302 and V303) are important in binding.

Relative stabilities of the LIM complexes

Having confirmed that both Lhx3 and Isl1 bind the same domain on Ldb1 and having determined that both Ldb1 and Isl1 bind the same site on Lhx3, we next sought to measure binding affinities for these different competing interactions to establish which complexes would be likely to form in vivo. Because the LIM domains of Isl1 and Lhx3 tend to be insoluble and prone to aggregation, it is not possible to measure the affinities using typical biophysical approaches. Thus, for the Ldb1/Lhx3 and Ldb1/Isl1 interactions, we generated tethered versions of the LIM–LID complexes where the linker contained a Factor Xa protease site (Figure 4A, inset). The tethered constructs were produced as GST fusion proteins and the linkers cut with Factor Xa to yield stable intermolecular LIM/LID and LIM/LBD complexes. FLAG-tagged Ldb1_LID was then used to compete off the unlabelled peptide in a competition ELISA, yielding dissociation constants of 35 nM for Lhx3/Ldb1_LID and 90 nM for Isl1/Ldb1_LID (Figure 4A).

Figure 4.

Binding affinities and modelling of complex formation. (A) ELISAs showing the ability of FLAG-tagged Ldb1_LID to bind to the cut Isl1/Ldb1_LID and Lhx3/Ldb1_LID complexes. Data were corrected for background binding to GST, and the error bars show the range of values obtained from triplicates of a single experiment. The inset shows a schematic of tethered intramolecular constructs of the Isl1–Ldb1_LID, where the linker contains a Factor Xa protease site. After treatment with Factor Xa, the linker is cut to form an intermolecular complex. (B) GdnHCl denaturation curves for Lhx3–Ldb1_LID and Lhx3–Isl1_LBD constructs as indicated. (C–E) Trends for the formation of protein and protein:DNA complexes containing Ldb1, Lhx3 and Isl1. Concentrations of all components were set at 1, 10 and 100 nM. (C) Distribution of Ldb1/LIM-HD complexes using the binary model (no interaction between Isl1 and Lhx3, and no DNA). (D) Distribution of protein-only complexes using the ternary model, which also considers binding of Isl1 and Lhx3. (E) Distribution of key protein:DNA complexes relative to Ldb1/Lhx3/DNA. The I series shows the results of using the independent model, the SA series shows the results of the single-molecule DNA-binding advantage model where the binding of two and four HDs to a single molecule of DNA shows increased binding affinities over single HD binding.

The competition ELISA approach used above did not yield good quality data for Lhx3_LIM1+2–Isl1_LBD, so we instead compared the relative stability of Lhx3–Ldb1_LID and Lhx3–Isl1_LBD constructs to chemical denaturation monitored by tryptophan fluorescence (Figure 4B). The Lhx3 domains and the linkers in each case are identical, thus any differences in the stability of the different complexes should correlate with relative differences in binding of the LIDs. We used constructs of tethered complexes in both orientations (i.e., LID–LIM and LIM–LID) to control for any differences in stability conferred by the position of the linker (Jeffries et al, 2006): these differences were very small for either pairing. Importantly, it was evident that the Isl1_LBD-containing complexes were significantly less resistant to GdnHCl denaturation than the Ldb1_LID-containing complexes (midpoint of denaturation ∼3.4 versus 5.8 M GdnHCl, respectively). Thus, although it was not possible to determine dissociation constants by this approach, Isl1_LBD appears to bind Lhx3 with significantly lower affinity than does Ldb1_LID.

Estimating the population distributions of Ldb1-, Lhx3- and Isl1-containing complexes

We then used our binding data and the program DynaFit3 (Kuzmic, 1996) to model the relative populations of the various binary and ternary complexes containing Ldb1, Lhx3 and Isl1, together with DNA sequences containing either single or double sites for either or both the HDs of Lhx3 and Isl1 (Figure 4C–E and Supplementary data 7). As the effective local concentrations of these proteins within the nucleus are unknown, this modelling was designed to assess trends over a range of different conditions rather than to predict actual concentrations of complexes. Protein concentrations in the range 1 nM up to 1 mM were used, taking into account both the predicted concentrations of nuclear transcription factors (∼1–100 nM) (Ryan et al, 2007) and the possibility of local concentration effects, which might give rise to much higher effective concentrations. Equal concentrations of all starting components were used.

If we just consider the competition of Lhx3 and Isl1 for Ldb1 (i.e., ignoring both Isl1_LBD and DNA; Figure 4C), simulations carried out at different concentrations reveal that in the absence of any other interactions, Ldb1_LID exhibits a preference for binding Lhx3 over Isl1 at the level of approximately two-fold more Lhx3/Ldb1 rather than Isl1/Ldb1 complexes being formed (e.g., 2.5-fold at 1 nM protein concentrations and 1.9-fold at 100 nM protein concentrations).

If we additionally consider the Lhx3–Isl1 interaction in our model (Figure 4D), we find that the ternary Lhx3/Isl1/Ldb1 complex is unlikely to be significantly populated. Even in the presence of DNA (Figure 4E), ternary complex formation (right-hand bars in Figure 4E) is favoured only if the four HDs in the ternary complex can bind simultaneously to multiple DNA sites (SA series) and gain a free energy ‘bonus' through the chelate effect (the reduction of entropy loss when binding a second site on the same molecule; Figure 4E).

The other situation in which the ternary complex is favoured, even in the absence of DNA, arises when the protein concentrations are set to be very high (mM concentrations; Supplementary data 7). Finally, the modelling data also indicate that significant levels of Lhx3–Ldb1 and Isl1–Ldb1 complexes are also likely to be present in cells in which Lhx3, Isl1 and Ldb1 are all coexpressed.

Discussion

The formation of cell-specific LIM-HD-containing complexes

The data from this study reveal that Isl1 is able to alter the nature of transcriptional complexes formed by Lhx3 and Ldb1 at neuronal promoters by binding Lhx3 through a decoy peptide domain within its own C terminus and allowing Isl1 to instead bind Ldb1_LID. The subsequent formation of the ternary Ldb1/Isl1/Lhx3 complex appears to be driven by enhanced binding to DNA sequences that contain multiple HD-binding sites. In support of this model, only constructs that can form ternary complexes (or artificial analogues of ternary complexes) with intact HDs from both Lhx3 and Isl1 can trigger motor neuron differentiation in cell lines (Thaler et al, 2002). Moreover, the enhancer for a key target of the ternary Ldb1/Isl1/Lhx3 complex in motor neuron development, Hb9 (Thaler et al, 1999), is bound only when both Lhx3 and Isl1 are present; neither Lhx3 nor Isl1 alone can bind (Lee and Pfaff, 2003). Thus, the ternary complex appears to be able to target different sets of genes compared with binary Ldb1/Lhx3 or Ldb1/Isl1 complexes. It is also possible that Ldb1/Lhx3 and Ldb1/Isl1 complexes are recruited to the vicinity of the motor neuron sites by interactions between Ldb1 and other transcription factors (Lee and Pfaff, 2003).

The common occurrence of Isl1/Lhx3 complexes

Isl1_LBD is a common feature in Isl proteins from complex organisms (Supplementary data 8). Residues in the spacer region can vary considerably, whereas residues within the binding motifs tend to be highly conserved. Lhx3 family proteins are also found in the same organisms, suggesting that the Isl/Lhx3 interaction is highly conserved and has been central to nervous system development from its evolutionary beginnings. It should be noted that the interaction of Lhx3 and Isl1 is not restricted to V2 interneurons and/or motor neurons; the two proteins interact in pituitary cells to regulate expression of the gonadotropin-releasing hormone receptor (Granger et al, 2006). Notably, the promoter of the follicle-stimulating hormone β gene contains at least six HD-binding elements that might be recognised by ternary complexes (West et al, 2004).

A comparison of LIM–Ldb1_LID complexes

We have previously determined structures of LIM domains from LMO proteins in complex with Ldb1_LID (Deane et al, 2003, 2004; Jeffries et al, 2006). The relative orientations of the LIM domains from tandem LIM structures of LMO4 and Lhx3 differ, but the LIM1 and LIM2 domains are very similar, despite small gaps within the sequence alignments of these LIM domains (Figure 5A and Supplementary data 6). Although equivalent residues in the LIM domains contact Ldb1_LID, the sequence conservation of those residues varies (Figure 5A). However, apart from significant structural differences around the hinge/spacer regions, there is a high level of structural conservation at the LIM/LID interfaces (Figure 5B and C). Mutagenic screens of Ldb1_LID versus four different LMO and LIM-HD proteins (Deane et al, 2004; Ryan et al, 2006; this study) implicate similar residues in Ldb1_LID as important for binding (Figure 5D). Several guiding principles emerge for Ldb1_LID/LIM interactions: (i) Ldb1_I322 is the key residue for binding LIM1; (ii) Ldb1_V303 is the key residue for binding LIM2 with a secondary cluster of residues around Ldb1_L309; (iii) mutations in both halves are often required to abrogate binding; and (iv) there is little or no contribution of residues from the spacer region in Ldb1_LID to binding, indicating that Ldb1_LID binds the tandem LIM domains of both LIM-HD and LMO proteins through two closely spaced binding motifs (Figure 5D). Thus, apart from different relative orientations of the two halves of the complexes, it is likely that Ldb1_LID binds all LMO/LIM-HD proteins in fundamentally the same manner.

Figure 5.

Comparison of LIM–Ldb1_LID complexes. (A) Structure-based sequence alignment of the LIM1 domains from murine LMO2, LMO4 and Lhx3 (Lhx3b numbering is used although the LIM domains are the same in different splice variants of Lhx3). Dots show conserved residues, dashes indicate gaps in the alignment and zinc-ligating residues are indicated by an asterisk. A consensus sequence for LIM domains from LIM-HD and LMO proteins from Deane et al (2004) is also shown. Fully conserved residues are indicated as capitals, + indicates R or K, − indicates D or E, and ϕ indicates bulky hydrophobic residues. Residues that are conserved at the interface of LIM–Ldb1_LID interactions are shaded (grey—hydrophobic; blue—backbone–backbone hydrogen bonds; yellow—other hydrogen bonds; red—electrostatic; many residues coloured blue, yellow or red also make hydrophobic interactions). Underlined residues in Lhx3 are at the interface of the Lhx3/Isl1_LBD complex. (B, C) Ldb1_LID from complexes containing LMO2 (1J2O, cyan), LMO4 (1M3V, violet; IRUT, magenta) and Lhx3 (yellow) shown in relation to the LIM domains from Lhx3 (grey surface representation) structures are overlaid over the backbone atoms of the LIM domains. (B) LIM1 domains/C-terminal half of Ldb1_LID as backbone traces. (C) LIM2 domains/side-chain heavy atoms of the N-terminal half of Ldb1_LID. (D) Summary of alanine mutagenic screening of Ldb1_LID against LMO2, LMO4, Lhx3 and Isl1. The sequence of Ldb1_LID is shown at the top; coloured boxes indicate a weak (yellow), moderate (orange) and strong (red) effect on binding when residues were individually mutated to alanine. Grey boxes indicate no differences. Where structures have been determined (LMO2_LIM1, LMO4, Lhx3), β-strands formed in Ldb1_LID are shown as black bars. The spacer region from the Lhx3–Ldb1_LID complex is shown as a green bar.

A mechanism for exchanging partners

The modular nature of LIM–peptide interactions, the apparent flexibility of the Ldb1_LID spacer and the observation that the Isl1_LBD can adopt different conformations (Figure 2E) all suggest a molecular mechanism for the disruption of the preferred Lhx3/Ldb1 complex by Isl1 (Figure 6). Lhx3 bound to Ldb1_LID (Figure 6A) makes stronger contacts through Lhx3_LIM2. It is possible that Lhx3_LIM1 periodically becomes ‘unstuck' whereas Lhx3_LIM2 remains in contact with Ldb1_LID (Figure 6B). When Isl1 is introduced to this half-complex, it is free to bind to Ldb1_LID through its favoured half (Isl1_LIM1), leaving Lhx3_LIM2 bound to the N-terminal half of Ldb1_LID (Figure 6C). Isl1_LBD would now be in close proximity to the Lhx3 LIM domains, encouraging the formation of an Isl1_LBD/Lhx3_LIM1 interaction (Figure 6D). Although this interaction is likely to be weak, it would be enhanced by the chelate effect. A final rearrangement of the LIM–LID interactions (Figure 6E and F) would give rise to the Ldb1/Isl1/Lhx3 complex.

Figure 6.

Putative mechanism for cofactor exchange between Lhx3 and Isl1. Only Ldb1_LID is shown for simplicity. (A) Lhx3 and Ldb1 form a complex where Ldb1_LID extends across both LIM domains of Lhx3. (B) Lhx3_LIM2 binds more tightly than Lhx3_LIM1 and for some of the time Lhx3_LIM1 is exposed while Lhx3_LIM2 remains bound. (C) Isl1_LIM1 can bind the C-terminal half of Ldb1_LID. (D) The C-terminal half of Isl1_LBD is brought into close proximity to Lhx3 and binds Lhx3_LIM1 (E, F) The N-terminal half of Ldb1_LID dissociates from Lhx3_LIM2 to bind Isl1_LIM2, allowing the N-terminal half of Isl1_LBD to bind Lhx3_LIM2. Complexes in (C–F) are stabilised through the simultaneous binding to both Lhx3 and Isl1 recognition sites on target genes.

Tandem binding motifs as mediators of protein signalling networks

Many protein:protein interactions are regulated through intrinsically unstructured motifs that take up a defined structure on binding to their partner (Dyson and Wright, 2005). More than 30% of protein sequence within eukaryotes is predicted to be intrinsically unstructured (Ward et al, 2004), and these regions exhibit a preponderance of short repeated sequences that vary in length and number of repeats (Tompa, 2003). The Lhx3-binding domains characterised here resemble tandemly arrayed linear motifs, although they lack the sequence conservation through which linear motifs have thus far been identified (Neduva and Russell, 2006). Given our observations of tandem binding repeats with very low levels of sequence homology, we suggest that tandem binding events between unstructured domains and folded protein domains could be extremely common.

Despite the plethora of newly defined unstructured domains in the literature, the majority of unstructured domains are yet to be characterised. For example, LIM-HD proteins (and indeed the majority of transcription factors) all contain long stretches of putatively unstructured protein sequence. We predict that other interactions mediated by intrinsically unstructured domains within LIM-HDs will have significant roles in establishing the identity of developing cells through the implementation of transcriptional LIM codes. Thus, it is likely that tandemly arrayed interaction motifs competing for binding to target proteins will emerge as an important general strategy for the formation and dynamics of multiprotein complexes.

Materials and methods

Cloning, mutagenesis and protein expression

All clones and mutants were generated by PCR and sequenced to confirm identity (SUPAMAC, Royal Prince Alfred Hospital, Sydney). Proteins were expressed with a GST tag using pGEX vectors (either pGEX-2T or a modified version of this vector in which the thrombin protease site was removed) in Escherichia coli BL21(DE3) at 20 or 25°C for ∼16 h. All proteins were purified by GSH-Sepharose 4B™ (GE Healthcare) affinity chromatography. Tethered proteins used for structural studies were additionally treated with thrombin to remove the GST tag and purified using either anion exchange chromatography (20 mM Tris–HCl, 1 mM DTT (pH 7.5 or 8.0) plus 0–1 M NaCl) or size-exclusion chromatography (Sephadex 75 20/60; GE Healthcare).

NMR solution structure determination and refinement

Data for chemical shift assignments for Lhx3_LIM1+2–Ldb1_LID were collected using the suite of triple resonance experiments as described previously (Lee et al, 2005). NOE distance restraints were obtained from three-dimensional ¹⁵N-NOESY-HSQC (80 ms mixing time) and ¹³C-NOESY-HSQC spectra (120 ms mixing time). Dihedral restraints were derived from backbone chemical shifts using TALOS (Cornilescu et al, 1999). Iterative manual assignment of NOEs was used to calculate the initial structures of Lhx3_LIM1+2–Ldb1_LID using ARIA (version 1.2) (Nilges et al, 1997). Of the structures calculated by ARIA, the best (lowest energy) 10 structures were used as template structures for further refinement in CNS (version 1.1) (Brunger et al, 1998). The quality of the ensemble of 20 best structures was assessed using PROCHECK-NMR (Laskowski et al, 1996). Chemical shift assignments, and NOE and dihedral restraint data have been deposited at BMRB (accession code 6658) and the coordinates for the ensemble have been deposited in the Protein Data Bank (2JTN).

X-ray structure determination and refinement

The crystallisation, collection and processing of a native data set for Lhx3_LIM1+2–Isl1_LBD were described previously (Bhati et al, 2008). Multiple wavelength anomalous dispersion data were recorded at three wavelengths near the zinc absorption edge: λ₁, 1.282 Å (the peak); λ₂, 1.170 Å (a high energy remote); λ₃, 1.283 Å (the inflection) on the GM/CA-CAT beamline 23ID-D at the Advanced Photon Source, Argonne National Laboratory, using a CCD detector, MARmosaic 300 (Marresearch). The synchrotron data were processed and scaled with HKL2000 (Otwinowski and Minor, 1997).

The zinc positions were located and the initial phases to 2.3 Å resolution were calculated with the program SOLVE (Terwilliger and Berendzen, 1999). The program RESOLVE (Terwilliger, 2000) was then employed for statistical density modification, local pattern matching and automated model building to 2.3 Å resolution. The resulting phases were combined with the high-resolution native data set and further automatic model building with phase extension to 2.05 Å resolution was carried out using ARP/warp (Perrakis et al, 1999). Manual manipulation of the structure was performed in COOT (Emsley and Cowtan, 2004). All refinement steps were carried out using REFMAC5 with TLS refinement (Murshudov et al, 1997; Winn et al, 2001). Analysis and validation of the structure were carried out with the assistance of the program PROCHECK (Laskowski et al, 1993) and the MOLPROBITY server (Lovell et al, 2003). The coordinates for the structures have been deposited in the Protein Data Bank (2RGT).

Yeast two-hybrid analysis

Y2H analysis was carried out using inserts cloned into modified pGBT9 and pGAD10 plasmids, and transformed into AH109 cells (Clontech), as described previously (Deane et al, 2004; Ryan et al, 2006). Selection conditions were all based on media lacking leucine and tryptophan to ensure co-transformation of both plasmids. For the detection of an interaction, media were also deficient in histidine and supplemented with X-α-Gal (40 μg ml⁻¹), further supplemented with 3-amino-1,2,4-triazole (3-AT; moderate stringency), or were deficient in both histidine and adenine (high stringency) as indicated.

Binding and denaturation experiments

Competition ELISA experiments (Deane et al, 2004) and GdnHCl denaturation experiments were performed as described previously (Jeffries et al, 2006). Unfolding was monitored by the tryptophan fluorescence wavelength maximum (λ_max) in the range 325–370 nm and data are reported as fraction folded. For the Lhx3–Ldb1_LID proteins, it was assumed that the λ_max of the fully unfolded proteins was 360 nm. Data were fitted to pseudo-two-state unfolding models to estimate the midpoints of denaturation of the curves (Fersht, 1998).

Modelling of population distributions of complexes

The relative populations of complexes that form were modelled using the program DynaFit3 (Biokin Inc., USA) (Kuzmic, 1996). For protein-only complexes, two different models were used: binary (B, no interaction between Isl1 and Lhx3) and ternary (T; interaction between Isl1 and Lhx3 was set at 200 nM). For protein binding in the presence of DNA, three different models were used: independent (I, each DNA site was considered to bind in an independent fashion with K_d=100 nM and was specific for either Lhx3 or Isl1), single-molecule DNA binding (S, protein complexes that bind two DNA sites were considered to bind with K_d=1 nM) and the single-molecule DNA-binding advantage (SA, four sites were additionally considered to bind with K_d=0.1 nM). The total concentrations of each protein and DNA species were all assigned the same value ranging from 1 nM to 1 mM in different simulations. For additional details, including the equations used in the models, see Supplementary data 7. The populations of key species under various conditions were estimated and are shown as relative populations (to Ldb1:Lhx3 or Ldb1:Lhx3:DNA).

Supplementary Material

Supplementary Figures