Solution Structure of the LIM-Homeodomain Transcription Factor Complex Lhx3/Ldb1 and the Effects of a Pituitary Mutation on Key Lhx3 Interactions

Lhx3 is a LIM-homeodomain (LIM-HD) transcription factor that regulates neural cell subtype specification and pituitary development in vertebrates, and mutations in this protein cause combined pituitary hormone deficiency syndrome (CPHDS). The recently published structures of Lhx3 in complex with each of two key protein partners, Isl1 and Ldb1, provide an opportunity to understand the effect of mutations and posttranslational modifications on key protein-protein interactions. Here, we use small-angle X-ray scattering of an Ldb1-Lhx3 complex to confirm that in solution the protein is well represented by our previously determined NMR structure as an ensemble of conformers each comprising two well-defined halves (each made up of LIM domain from Lhx3 and the corresponding binding motif in Ldb1) with some flexibility between the two halves. NMR analysis of an Lhx3 mutant that causes CPHDS, Lhx3(Y114C), shows that the mutation does not alter the zinc-ligation properties of Lhx3, but appears to cause a structural rearrangement of the hydrophobic core of the LIM2 domain of Lhx3 that destabilises the domain and/or reduces the affinity of Lhx3 for both Ldb1 and Isl1. Thus the mutation would affect the formation of Lhx3-containing transcription factor complexes, particularly in the pituitary gland where these complexes are required for the production of multiple pituitary cell types and hormones.


Introduction
Lhx3 (LIM homeobox protein 3) is essential for specification of many pituitary and neural cell types [1,2,3]. Humans that carry mutations in Lhx3 present with combined pituitary hormone deficiency syndrome (CPHDS) [4,5,6,7,8]. Depending on the site of mutation, affected patients can also exhibit hearing loss and skeletal malformations of the upper body [8,9].
Lhx3 is from the LIM-homeodomain transcription factor family, members of which contain a pair of closely spaced Nterminal LIM domains followed by a central homeodomain ( Figure 1A). The C-terminal portion of LIM-homeodomain proteins is generally poorly characterized but in Lhx3 is reported to contain an activation domain [10]. LIM domains (named for the first three proteins in which the domain was found, Lin-11, Isl1 and Mec-3) are zinc fingers that coordinate two zinc ions and mediate protein-protein interactions [11]. The LIM domains from Lhx3 make well-characterized interactions with Ldb1 (LIM domain binding protein 1) and Isl1 (Islet 1) [2,12,13,14,15] and have been reported to also bind PIT-1 (pituitary-specific transcription factor 1) [16] and SLB (selective LIM domain binding protein) [17]. The DNA-binding specificity of Lhx3 is context dependent, and varies according to the protein isoform [18,19], or whether Lhx3 is in acting in concert with a protein partner [2,20,21,22]. Disease-causing mutations or post-translational modifications, including phosphorylation, of the LIM domains are likely to affect the biological activity of Lhx3 by modulating protein-protein interactions and modulating binding to DNA targets [5,6,10].
The isolated LIM domains from Lhx3 (Lhx3 LIM1+2 ) tend to be insoluble and/or aggregate, but soluble stable ''tethered complexes'' can be engineered in which the LIM interaction domain of Ldb1 (Ldb1 LID ), or the Lhx3-binding domain from Isl1 (Isl1 LBD ) are fused to Lhx3 LIM1+2 via a flexible glycine-serine linker ( Figure 1B) [23,24,25]. These tethered complexes are hereafter referred to as Ldb1-Lhx3 and Lhx3-Isl1 (the order of the names indicates the order of the domains in the complex). Our recently determined structures of Ldb1-Lhx3 and Lhx3-Isl1 show that Ldb1 LID and Isl1 LBD interact with Lhx3 in an essentially identical manner [13], with the two binding partners forming extended chains and contacting the same sites across both LIM domains of Lhx3. The tethered complexes appeared to have some flexibility in the ''hinge'' between the LIM domains and the corresponding ''spacers'' in Ldb1 LID and Isl1 LBD which lie between the two LIMbinding motifs in each of those domains. The hinge comprises Lhx3 F90 , which shows some variation in backbone angle in different conformers/molecules, and the spacers comprise Ldb1 M310-D318 and Isl1 H273-Q278 , which assume different overall conformations in the Ldb1 and Isl1 structures (including a short region of disorder in Ldb1 G312-G315 and an extended structure in one chain of Isl1, but a turn in the other chain) [13]. For Lhx3-Isl1 flexibility at the hinge/spacer is supported by small angle X-ray scattering (SAXS) data [26]. The initial part of this study uses SAXS to further characterize the solution structure of Ldb1-Lhx3.
These structures provide us with the opportunity to interpret the molecular effects of disease-causing mutations and posttranslational modification of the LIM domains of Lhx3. The mutation of tyrosine 111 to cysteine, Y111C, is an inherited point mutation found in the LIM domains of human Lhx3 that is associated with CPHDS [4]. Although the sequences of the LIM domains from Lhx3 are almost identical in mammals, the numbering of the human and mouse proteins differs slightly; numbering for the mouse protein is used herein, with human Y111 corresponding to Y114 in the mouse protein. The affected residue lies adjacent to one of the zinc ligating residues, H115, prompting suggestions that the introduced cysteine sidechain in Y114C might displace H115 as a zinc ligand [5,6]. Two putative phosphorylation sites, T63 and S71 are located in the first LIM domain of human Lhx3 [10], of these residues only S71 is highly conserved across species. The equivalent residue in the mouse protein, S74, lies adjacent to the binding interface within the Lhx3/ Ldb1 and Lhx3/Isl1 complexes.
Here we used SAXS to show that the NMR structure of Ldb1-Lhx3, an ensemble of elongated molecules with small differences in angle between the two LIM modules [13], is a reasonable representation of the solution structure of this tethered complex.
We used yeast two-hybrid assays to show that pseudophosphorylation of Lhx3-S74 does not affect binding to Ldb1 or Isl1, but that Lhx3(Y114C) has reduced levels of binding to both key partners. NMR spectroscopy and stability studies demonstrated that Lhx3(Y114C) does not alter the zinc-ligation characteristics of this domain, but does affect the stability and local structure of the second LIM domain of Lhx3.
Yeast Two-hybrid Assays pGBT9 and pGAD10 plasmids were co-transformed into AH109 cells (Clontech), as described previously [27]. All selective media lacked leucine (-L) and tryptophan (-W) to ensure cotransformation of bait and prey plasmids was maintained. For screening of interactions, all media were further deficient in histidine (-H) but contained or lacked additional reagents for selection of different affinity interactions. Selective media were either supplemented with 40 mg mL 21 X-a-gal (Progen) and 1 mM 3-amino-1,2,4-triazole (3-AT; Sigma) (-L-W-H +3-AT; moderate stringency selection), or were additionally deficient in adenine (-L-W-H-A; high stringency selection).

Circular Dichroism Spectropolarimetry
Far UV-CD experiments were recorded at 20uC on a Jasco J-720 spectropolarimeter equipped with a Neslab RTE-111 temperature controller. Protein samples (30 mM) were prepared in 20 mM Na 2 HPO 4 , 40 mM NaCl, 1 mM DTT, pH 6.8, placed in a 1-mm path length quartz cell seated in a water-jacketed cell holder. CD spectra were collected over the wavelength range 205-260 nm, a speed of 20 nm/min, 1-nm step resolution, 1-nm bandwidth and a response time of 1 s. Final spectra were the average of five scans, and were buffer baseline corrected. Estimates of secondary structure were determined using CDPro [28].

Chemical Denaturation Experiments
Protein (2.5 mM) in 20 mM Na 2 HPO 4 , 40 mM NaCl, 1 mM DTT, pH 6.8 and Gdn.HCl as indicated were incubated at 25uC for 2-3 h. Fluorescence emission spectra (320-370 nm) were recorded using a Varian Cary Eclipse fluorescence spectrometer (Palo Alto, CA, USA), with an excitation wavelength of 295 nm. Slit widths were 10 nm, data were collected in 1-nm steps with an averaging time of 0.5 s, and spectra were buffer baseline corrected.
Data were processed using XWINNMR 3.5 or TOPSPIN 1.2 (Bruker Biospin). Spectral analysis was carried out using SPARKY versions 3.05-3.113 [30]. 1 H frequencies of all spectra were directly referenced to DSS, and 15 N frequencies were referenced indirectly [31]. Spectral resolution in the directly detected dimension was enhanced by apodisation with either a Lorentzian-Gaussian window function (LB = 0.1, GB = -3) or a squared sine bell function (shifted by p/2.6 radians) in indirectly detected dimensions. Digital resolution was enhanced by zero-filling (once in each dimension) and linear prediction (in the 15 N dimension only) before Fourier transformation. Polynomial baseline corrections were applied to processed spectra where appropriate.

Small Angle X-ray Scattering
Small-angle X-ray scattering data were collected from solutions of Ldb1-Lhx3 (4.5-9 mg mL 21 ) with 20 mM Na 2 HPO 4 , 40 mM NaCl, 1 mM DTT, pH 6.8 and a matched solvent blank at 283 K for 10 s intervals over 30 min using a line-collimated SAXSess scattering instrument (Anton Paar, Graz, Austria) equipped with a CCD detector [32]. Scattering data were reduced to I(q) vs q (where q = (4psinh)/l, 2h is the scattering angle, and l the X-ray wavelength, CuK a , 1.54 Å ) using the SAXSQuant 2.0 software package (Anton Paar, Austria) that corrects for sample absorbance and detector sensitivity, and normalises and subtracts solvent from protein+solvent to yield I(q) versus q for the protein alone. The reduced scattering profiles were all placed on an absolute scale using the scattering from water [33]. The programs GIFT [34] and GNOM [35] were used to calculate the probable distribution of atom-pair distances within Ldb1-Lhx3 (P(r) versus r profiles), accounting for the 10-mm slit-geometry of the instrument, from which the radius of gyration (R g ), maximum dimension (D max ) and forward scattering intensity at zero angle (I(0)) were extracted; the two programs gave essentially the same results. A molecular weight estimate of Ldb1-Lhx3 was derived from I(0) as described in [33] using values for the contrast (Dr) and partial specific volume (u) calculated in CONTRAST from the MULCh program suite [36]. Although MULCh was designed for use in small-angle nuclear scattering (SANS), it can also be used to process SAXS data. An I(0) analysis of the 9 mg mL 21 sample was indicative of some aggregation and was not further analysed. Guinier analysis of the SAXS data was performed using PRIMUS [37]. Ab initio shape restorations of Lhx3-Ldb1 were performed 10 times using the program DAMMIF [38] and a consensus model developed via the spatial alignment and averaging of each solution combined with standard phase-occupancy and volume corrections [39]. Rigidbody modelling against the SAXS data was performed using the high resolution structures of each LIM-half of the complex and the program BUNCH [40]. CRYSOL [41] was used to evaluate the fits against the scattering data of the final BUNCH model and each individual Ldb1-Lhx3 NMR structure derived from the NMR ensemble. CRYSOL was also used to calculate the theoretical scattering profiles of the resultant high-resolution models which were used to derive model P(r) vs r profiles for comparison with the experiment [35].

SAXS Data are Consistent with NMR Data for Ldb1-Lhx3
Our previously determined NMR structure of Ldb1-Lhx3 is elongated, with members of the NMR ensemble comprising two LIM modules (LIM1 and LIM2, each with the contacting region of Ldb1 LID ) but with angles at the ''hinge''/''spacer'' between the two modules that vary by up to ,30u, and the tether between Ldb1 and Lhx3 and residues at the C-terminus of the construct are unstructured (Figure 2A and B) [13]. Similar tethered complexes (Lhx3-Isl1 and Lhx4-Isl2) can give rise to extreme angles in crystal structures [13,26]. Relatively few long distance restraints exist between the two modules in the NMR structure, and residual dipolar coupling constants could not be determined for this complex [13], creating some uncertainty as to whether the gross structure of the Lhx3/Ldb1 complex was adequately described. Thus we used small-angle X-ray scattering (SAXS) to independently define the global conformation of Ldb1-Lhx3 in solution. Analysis of the molecular weight derived from the forward scattering intensity at zero angle combined with the linearity of the Guinier plots in the low-q regime indicates that Ldb1-Lhx3 exists as a monodisperse sample of monomers (Table 1; Figure 2C) [42].
The experimentally derived P(r) profile ( Figure 2D) is characteristic of an elongated molecule [43,44], and is consistent in overall dimensions (,90620 Å ; D max = 88 Å ; R g = 25.3 Å ) with members of the NMR ensemble for Ldb1-Lhx3 (R g NMR = 24-25.1 Å ; Figure 2A and B, Table S1). The 20 conformers of the NMR ensemble show a range of fits to the SAXS data, with x 2 values ranging from 0.91 to 1.37 (Text S1, Table S1). The conformer that best fits the data is Model 17 ( Figure 2C and D), which still shows a small deviation from the experimental SAXS data (e.g. Figure 2D).
To generate models of Ldb1-Lhx3 based solely on SAXS data, ab initio shape reconstructions of Ldb1-Lhx3, which are independent of high-resolution model bias, were calculated using DAMMIF. Each of the ten individual solutions used to generate the consensus DAMMIF shape of the complex ( Figure 2E) fit the experimental data well (x 2 of 0.7) and have an average normalised spatial discrepancy of 0.63 indicating that the solutions used to generate the consensus model are similar [39]. We also performed rigid-body modelling of Ldb1-Lhx3 using BUNCH [40] in which the positions of the two LIM-modules of the complex were allowed to flex relative to each other during refinement against the SAXS data, as described for related Lhx3/4-Isl1/2 complexes in [26]. The resultant BUNCH model fits the SAXS data with x 2 = 0.75. Model 17 from the NMR ensemble, the BUNCH model and the DAMMIF shape reconstruction all superimpose well ( Figure 2E). The differences between the BUNCH and NMR models are some twisting at the hinge/spacer, and repositioning of the unstructured linker, such that it is more compact in the BUNCH model. By definition, the positions of atoms in unstructured regions of the NMR ensemble are inadequately defined, so although the unstructured residues are present in each conformer their given coordinates are essentially arbitrary. Swapping linkers between models from the NMR ensemble to a more compact linker can improve the fit to the SAXS data, suggesting that inadequate modelling of the linker is a major contributor to the poor fits of some NMR conformers (Text S1, Figure S1, Table S2). Indeed, the structured regions (but not the undefined tethers) of the NMR ensemble superimpose well the DAMMIF consensus model ( Figure 2F), indicating that the range of angles at the hinge/ spacer in the NMR ensemble ( Figure 2G) is reasonable.
Overall these data suggest that the NMR ensemble is a reasonable representation of the complex in solution: two relatively rigid LIM domains in an overall extended orientation with an interdomain angle that can vary by up to 30u.

Analysis of Interactions between Lhx3 Mutants
We used yeast two-hybrid analysis to test if the Lhx3(Y114C) mutant and two pseudo phosphorylation mutants [45], Lhx3(S74D) and Lhx3(S74E), affected the interaction of that protein with Ldb1 LID or Isl1 LBD (Figure 3). In these experiments the apparent strength of the interactions depends on which vector contains the Lhx3 construct; yeast growth is more robust when Lhx3 is fused to the activation domain in the pGAD10 vector than to the DNA-binding domain in pGBT9 [13]. No interaction was observed between Lhx3(Y114C) and Isl1 LBD under any selection conditions, whereas yeast growth was observed under moderate (-L-W-H +3-AT) but not high stringency (-L-W-H-A) selection conditions for binding to Ldb1 LID when Lhx3 is in the pGAD10 vector, indicating that the mutation abrogates binding to Isl1 LBD and significantly reduces binding to Ldb1 LID (Figure 3). The Lhx3(S74D) phosphomimic did not appear to affect either interaction, whereas Lhx3(S74E) moderately reduced binding to Isl1 LBD ; no yeast growth was seen under high selection conditions (-L-W-H-A) when Lhx3 was in the pGAD10 vector.

Characterisation of the Lhx3(Y114C) Mutant
A tethered Ldb1-Lhx3(Y114C) construct was generated to determine whether misfolding and/or non-native zinc ligation caused reduced binding by the Y114C mutant. Far-UV circular dichroism spectra of the wildtype and mutant proteins were very similar ( Figure 4A) indicating that the wildtype and mutant proteins have identical levels of secondary structure (6-7% ahelical structure, 33-35% b-structure, and 58-61% coil). The relative stability of the mutant and wildtype tethered complexes was assessed using resistance to chemical denaturation monitored by intrinsic tryptophan fluorescence as described previously [13,26,46]. Both complexes contain a single tryptophan in Lhx3 LIM1 , and their tryptophan fluorescence emission spectra show a red-shift of wavelength maximum and reduction in intensity typical of unfolding when exposed to 6 M Gdn.HCl (data not shown). Both proteins displayed a monophasic unfolding transition ( Figure 4B), but it should be noted that these proteins each contain two LIM domains and a binding peptide meaning We use the convention for reporting SAXS data as outlined in ref [42].  that the folding was unlikely to be two-state or fully reversible and free energies of folding could not be determined. The mutant tethered complex (midpoint of denaturation, D 50% = 2.0 M Gdn.HCl) was apparently destabilised relative to the wildtype complex (D 50% = 2.5 M Gdn.HCl). The stability of these complexes stems from both the intrinsic stability of the LIM domains and the affinity of the LIM domains for Ldb1 LID , and so the decrease in the overall apparent stability of the Ldb1 LID -Lhx3(Y114C) mutant likely reflects a weaker interaction and/or decrease in stability of the Lhx3 LIM1+2.
The structure of the mutant protein was further probed by NMR spectroscopy. The peaks in the 15 N-HSQC spectrum of uniformly labelled 15 N-Ldb1-Lhx3(Y114C) were sharp and well dispersed indicating that the protein is folded; however, many of the peaks have shifted compared to the spectrum for the wildtype protein ( Figure 4C); we have inferred assignments for the mutant protein from resonance proximity in two-dimensional spectra ( Figure S2 and Table S3). Although caution should be taken in the interpretation of these spectra because we have not unambiguously assigned the spectra, the majority of the peaks that have moved in the spectrum of the mutant protein correspond to residues in the Lhx3 LIM2 domain whereas the N-terminal half of Ldb1 LID and Lhx3 LIM1 appear to be largely unaffected by the mutation (Figure 4D, Figure S2 and Table S3).
We analysed the protonation pattern of the H115 sidechain using 15 N-HSQC experiments [29] to investigate the possibility that the Y114C mutation might alter the zinc coordination. The two histidine residues that ligate zinc ions in Ldb1-Lhx3, Lhx3(H55) and Lhx3(H115), give rise to patterns that are typical of protonation at N e2 , indicating that zinc ligation occurs through N d1 in both cases ( Figure 5). Any change in the zinc coordination state of those histidine sidechains would result in changes to both the pattern and intensity of the peaks in the spectrum of the mutant protein ( Figure 5A). Although there are some changes in the 15 N chemical shift resonances from the N d1 nitrogen of H115 in the spectrum from Ldb1-Lhx3(Y114C), the lack of movement of the resonances corresponding to the N e2 , H e1 and H d2 nuclei suggests that the ligation state of H115 is the same in both the wildtype and mutant proteins ( Figure 5B). That is, C114 in the mutant protein does not replace H115 as a zinc ligand. The movement of the N d1 nitrogen of H115 can be attributed to its proximity to the site of mutation; a bulky aromatic tyrosine sidechain has been substituted by cysteine, likely causing a change in the local fold and/or electronic structure of the protein.  15 N-HSQC spectra of Ldb1-Lhx3 (black, ,800 mM)) and Ldb1-Lhx3(Y114C) (magenta, ,200 mM) in 20 mM Na 2 HPO 4 , 40 mM NaCl, 1 mM DTT, pH 6.8 at 310K. (D) Analysis of chemical shift differences from panel C based on assignments for the wildtype protein [25] and inferred assignments for the mutant protein (Table S3). Peaks were identical for the C-terminal half of Ldb1 LID , and Lhx3 LIM1 which make direct contacts in the structure of the complex, but were significantly different for the N-terminal half of Ldb1 LID , and Lhx3 LIM2. doi:10.1371/journal.pone.0040719.g004

Discussion
The SAXS analysis of Ldb1-Lhx3 confirms that the previously published NMR structure of this tethered complex is a reasonable representation of its solution structure -an ensemble of conformers that varies in angle by up to ,30u between the two LIM modules. Given the extended nature of this complex and the paucity of contacts between the two LIM modules we expect that the complex undergoes a limited amount of flexion at this point and/ or twists as suggested by the BUNCH model ( Figure 2). The SAXS data for Ldb1-Lhx3 are consistent with a small amount of flexion, but not with high levels of flexibility. Previously reported heteronuclear 15 N-1 H NOE values are consistent with no or limited flexion between the LIM domains, although increased motion in residues Ldb1(E313-G315) corresponds with a short region of disorder in Ldb1 LID spacer [13]. Flexibility at the hinge/  spacer was suggested for a related LIM-only protein 2 (Lmo2)-Ldb1 LID complex [47], and is consistent with SAXS data for Lhx3/4-Isl1/2 complexes [26]. A phenylalanine residue followed by a glycine residue is highly conserved at this hinge in LIMhomeodomain and LIM-only proteins ( Figure S3), suggesting that some flexion at the hinge/spacer is a common phenomenon in this type of complex. Flexibility was suggested to play a role in the binding of Lmo2/Ldb1 to Tal1/E47 [47], and differences in the inter-LIM domain angle could influence the recruitment of additional binding partners to and thereby generate differences in activity between alternate Lhx3-containing complexes. We recently showed that the complex formed between LIM homeobox protein 4 (Lhx4) and Islet 2 (Isl2) and that formed between Lhx3 and Isl2 have a more compact average structure (D max ,75 Å ) than a similar complex formed by Lhx3 and Isl1 (D max ,90 Å ), suggesting that the binding partner can influence the average angle formed between the two LIM domains [26]. Despite a higher sequence identity between Isl1 LBD and Isl2 LBD than Isl1 LBD and Ldb1 LID , SAXS data for Ldb1-Lhx3 (which show, for example, a maximum dimension D max of 88 Å ) indicate that the gross structure of this complex resembles the more elongated Lhx3/4-Isl1 complexes. The spacing and structure of spacer may play an important role in determining the inter-LIM angle; the spacers in Isl1 LBD and Isl2 LBD have very low sequence identity compared to the LIM-binding motifs in those domains.
Although S74 lies proximal to the binding faces in both complexes ( Figure 6), neither of the phosphomimic mutants of Lhx3 had a major effect on binding to Ldb1 and Isl1 using yeast two-hybrid assays. Only S74E showed some apparent reduction in binding suggesting that the larger size (rather than the charge) of the glutamate sidechain was responsible for this effect. It is plausible that in Lhx3(S74E) local rearrangements in the LIM1 domain may be required to accommodate the larger glutamate sidechain resulting in a minor reduction of binding. These results are consistent with data from Parker et al. showing that Lhx3(S74A) mutants did not affect binding of Lhx3 to Ldb1, PIT-1 or MRG in GST-pulldown experiments [10]. If Lhx3 is phosphorylated at this site as part of normal activity it may be to modulate binding to as yet unidentified or uncharacterised protein partners.
Our yeast two-hybrid data using the CPHDS-associated mutation Lhx3(Y114C) confirm the reduced binding observed between Lhx3 mutants and Ldb1 in GST-pulldown experiments [6]. That the mutation also severely affects interactions with Isl1 is consistent with our structural and mutagenic scanning data showing that Isl1 and Ldb1 interact with the same site on Lhx3, with the majority of critical contacts being made at the LIM2 domain [13]. Y114 is highly conserved in the LIM domains of LIM-homeodomain and LIM-only proteins ( Figure S3) suggesting that Y114 is crucial for the structure and/or function of these proteins. Y114 does not contact either Ldb1 LID or Isl1 LBD ( Figure 6A), nor does the Lhx3(Y114C) mutation appear to induce a change in zinc ligation ( Figure 5). Rather, the highly buried sidechain points away from the Ldb1 LID and Isl1 LBD binding interfaces and forms a major part of the hydrophobic core of the first zinc binding module in Lhx3 LIM2 . Substitution of the bulky tyrosine sidechain with cysteine could result in the repacking of the hydrophobic core of Lhx3 LIM2 , which is consistent with the major chemical shift changes observed in the NMR spectrum of the Ldb1-Lhx3(Y114C) mutant and the reduced stability of Ldb1-Lhx3. However, Lhx3(Y114F) has a similar phenotype to Lhx3(Y114C) [5] suggesting that other factors may be at play. The hydroxyl group in the Y114 sidechain forms a hydrogen bond with the main-chain carbonyl from Lhx3(S140) (Figure 6), which would be lost on both mutants potentially destabilising the domains. Although interactions between Lhx3 and Ldb1 and Isl1 are required for motor neuron development [2], patients carrying the mutation do not show signs of motor neuron impairment [4]. It is likely that the closely related Lhx4 protein, which is also expressed in motor neurons [3], may compensate for mutant Lhx3 in developing motor neurons but not in pituitary development [48]. In motor neurons both Lhx3 and Lhx4 have high levels of expression, whereas only Lhx3 is highly expressed at multiple stages of development, but Lhx4 appears to have a much more restricted expression pattern (MGI database, http://www. informatics.jax.org/).
In conclusion, our analysis of Lhx3 complexes provides a structural framework with which to understand how Lhx3 is regulated in the cell, and how the biological functions of the protein may be affected by disease causing mutations. Figure S1 Analysis of SAXS data. (A) Models 1 (cyan linker), 8 (magenta linker), 9 (orange linker) and 17 (black linker) from the NMR structure of Ldb1-Lhx3 (pdb accession code: 2JTN) with Ldb1 LID in yellow and Lhx3 in blue; zinc ions are depicted as grey spheres. The models were aligned over the backbone atoms of Lhx3 LIM1 and the corresponding region of Ldb1. (B) Scattering data for Ldb1-Lhx3 (grey circles) shown as I(q) against q plots. Curves are the fits to the data generated by CRYSOL of sample NMR models from the Ldb1-Lhx3 NMR ensemble as indicated. Note that the scale on the Y-axis is arbitrary and the data/fits have been plotted separately for clarity. (EPS) Figure S2 Overlay of 15 N-HSQC spectra for wildtype and mutant Ldb1-Lhx3 constructs. 15 N-HSQC spectra of Ldb1-Lhx3 (black; 800 mM protein) and Ldb1-Lhx3(Y114C) (red; ,200 mM protein). Spectra were collected at 310 K in 20 mM Na 2 HPO 4 , 40 mM NaCl, 1 mM DTT, pH 6.8. Assignments for wildtype Ldb1-Lhx3 [25] are shown in black; for clarity, not all assignments are shown. Note that construct residues 1-45 refer to Ldb1 295-339 , construct residues 46-56 correspond to the synthetic linker, and construct residues 57-182 refer to Lhx3   (See  Table S3 for full list). (EPS) Figure S3 Sequence alignment of LIM domains from murine LIM-HD and LMO proteins. Zinc-coordinating residues are marked with an asterisk. Gaps in the sequence are indicated with a dash. Residues equivalent murine Lhx3 Y114 are highlighted in yellow. Residues equivalent to LMO2 F88/G89 are highlighted in cyan and Lhx3 S74 is highlighted in green. Consensus sequences are indicated in red. W, hydrophobic; +, positively charged; -, negatively charged; Z, aspartate or histidine. (EPS)  [25], the inferred assignments for the Y114C mutant and the weighted average chemical shift differences between those assignments. Note that not all peaks could be assigned by this approach. The shaded parts of the table refer to assignments in the N-terminal half of Ldb1 LID and Lhx3 LIM2. (DOCX)

Supporting Information
Text S1 Fits of NMR structures and SAXS data. (DOCX)