Structure of the mouse acidic amino acid decarboxylase GADL1

The structure of the mouse glutamic acid decarboxylase-like protein 1 (GADL1) is described. The structure provides new insights into the function of GADL1 and related decarboxylases.


Introduction
Pyridoxal 5 0 -phosphate (PLP), or vitamin B 6 , is a versatile cofactor that is involved in many enzymatic reactions spanning multiple enzyme classes and chemical reactions (Percudani & Peracchi, 2003). PLP-dependent decarboxylases constitute a large family of enzymes that catalyze a range of metabolic reactions. Many of these enzymes catalyze biologically well defined processes; inactivating mutations affecting them are associated with severe phenotypes, and some of them are treatment targets in human disease (Baekkeskov et al., 1990;Eliot & Kirsch, 2004;El-Sayed & Shindia, 2011;Paiardini et al., 2014;Skö ldberg et al., 2004).
Despite extensive research, the biological functions of many PLP-dependent enzymes are still unclear. One such enzyme is glutamic acid decarboxylase-like protein 1 (GADL1), which was originally named based on its sequence homology to glutamic acid decarboxylase (GAD), which synthesizes the inhibitory neurotransmitter -aminobutyric acid (GABA; Fenalti et al., 2007). However, GADL1 has no reactivity towards glutamic acid and therefore is unlikely to be involved in GABA signalling (Liu et al., 2012;Winge et al., 2015). It has been suggested that GADL1 is involved in taurine production (Liu et al., 2012). On the other hand, in our recent comparative study of GADL1 and cysteine sulfinic acid decarboxylase (CSAD), an enzyme homologous to GADL1 that is involved in taurine biosynthesis, we showed that these enzymes act differently. Compared with CSAD, the activity of GADL1 towards cysteine sulfinic acid (CSA) as a substrate is much lower, and GADL1 has a stronger preference for aspartate, suggesting that GADL1 may be involved in the biosynthesis of -alanine and its peptide derivatives, such as the neuroprotectant carnosine (Min et al., 2008;Park et al., 2014;Winge et al., 2015).
We showed in our earlier study that mouse and human brain have distinct patterns of expression of CSAD and GADL1 (Winge et al., 2015). Whereas both CSAD and GADL1 were expressed in neurons, only the CSAD mRNA was detected in astrocytes. Using a homology model of GADL1 based on the crystal structure of human CSAD (HsCSAD), we performed in silico screening of potential substrate analogues and discovered the first generation of inhibitors with partial selectivity against either GADL1 or CSAD. However, detailed mechanistic studies have been hampered by the lack of structural information.
In this study, we describe the crystal structure of mouse GADL1 (MmGADL1). Additionally, we used small-angle X-ray scattering (SAXS) to elucidate the solution shape of MmGADL1. The overall fold of MmGADL1 is similar to those of CSAD and other close homologues, with a flexible loop, not defined in electron density, from the apposing monomer covering the active site, which is possibly relevant in catalysis. SAXS data demonstrate that MmGADL1 adopts a loosened state in solution, which might correspond to an open conformation significant for cofactor or substrate binding.

Materials and methods
2.1. Macromolecule production 2.1.1. Construct preparation, protein expression and purification. The expression vector for MmGADL1 was prepared in the Gateway system using pTH27 (Hammarströ m et al., 2006) as the destination vector. Cloning involved a twostep PCR protocol and homologous recombination into Gateway vectors using standard procedures. The resulting construct codes for an N-terminal His 6 tag, a Tobacco etch virus (TEV) protease cleavage site and MmGADL1 residues 1-502 (UniProt entry E9QP13). A longer isoform of MmGADL1 also exists (UniProtKB entry Q80WP8), and the construct corresponds to residues 49-550 of this isoform.

Crystallization
Initial crystallization conditions for MmGADL1 were obtained from the JCSG-plus screen (Molecular Dimensions) using sitting-drop vapour diffusion. The crystallization conditions, which yielded crystals with a needle morphology arranged as single crystals or point-originated clusters, were comprised of 80 mM sodium cacodylate pH 6.0-7.4, 13-14%(w/v) PEG 8000, 120-160 mM calcium acetate, 15.0-17.5%(w/v) glycerol. 0.3-1.5 ml drops with different volume ratios of protein solution (6.5-7.5 mg ml À1 in 20 mM HEPES pH 7.4, 200 mM NaCl) and reservoir solution were used at 281 and 293 K, equilibrating against 40 ml reservoir solution. Crystals were briefly soaked in cryoprotectant solutions and flash-cooled in liquid N 2 . The detailed conditions used to obtain the crystals used for data collection are given in Table 1.

Data collection, structure solution and refinement
The MmGADL1 structure was solved from the two crystal forms by molecular replacement in Phaser (McCoy et al., 2007) using the structure of HsCSAD (PDB entry 2jis; Structural Genomics Consortium, unpublished work) as the search model. Crystal form 1 diffracted to 3 Å resolution, while crystal form 2, which was used for initial structure solution, diffracted to 2.4 Å resolution; the latter suffered from near-  perfect pseudomerohedral twinning and high translational pseudosymmetry. Owing to these observations, both crystal forms were solved and initially refined, and the lower resolution structure, which completely lacked twinning and pseudotranslation, was considered to be better for final refinement. In essence, despite the higher nominal resolution, the twinned crystal suffering from pseudotranslation gave lower-quality electron-density maps. The twinning and pseudotranslation operations are given in Table 2. NCS restraints were employed throughout refinement. Refinement was performed with phenix.refine (Afonine et al., 2012) and model building with Coot (Emsley & Cowtan, 2004). The structure was validated with MolProbity (Chen et al., 2010). Data collection and refinement statistics can be found in Table 2. The resolution limits used were based on recent studies (Karplus & Diederichs, 2015) showing that useful information is available for refinement even for data with a CC 1/2 much lower than 50%.

Small-angle X-ray scattering
Synchrotron SAXS data were collected on the EMBL/ DESY BioSAXS beamline P12 (Blanchet et al., 2015). The protein was at 1.6-6.5 mg ml À1 in 20 mM HEPES pH 7.4, 200 mM NaCl. The scattering data were processed and analyzed with programs from the ATSAS package (Konarev et al., 2006). The molecular weight was determined by comparison of the forward scattering intensity, I(0), with a fresh monomeric bovine serum albumin standard. Models of MmGADL1 were built with GASBOR  and SREFLEX (Panjkovich & Svergun, 2016), using data extrapolated to zero concentration. Theoretical scattering curves from crystal structure coordinates were calculated with CRYSOL (Svergun et al., 1995). Model superposition was performed using SUPCOMB (Kozin & Svergun, 2001). Details of SAXS data collection, processing and analysis are given in Table 3, and the raw SAXS scattering data are provided as Supporting Information.

Results and discussion
3.1. The crystal structure of MmGADL1 Initial screening for crystallization conditions of MmGADL1 resulted in a single condition that produced diffracting crystals. Crystals formed with a needle morphology, often growing in bunches or clusters and generally being very thin, with maximum lengths reaching 500 mm. The diffraction quality was initially poor, with diffraction limits of around 6-7 Å and highly smeared reflections. By optimizing the size, the amount of nucleation, and the general appearance of the crystals, the conditions given in Table 1 produced thin but nonfragile crystals that allowed structure refinement in two crystal forms to resolutions of 2.4 and 3.0 Å ; the higher resolution data set was plagued by significant twinning and pseudotranslation (Table 2). Thus, the structure of the nontwinned crystal form is discussed here; essentially all features can also be seen in the twinned crystal.
The crystal structure revealed a single MmGADL1 homodimer in the asymmetric unit, which was the expected oligomeric state based on other PLP-dependent decarboxylases (Fig. 1a). For both chains, residues 11-502 could be built, with the flexible loop common to PLP-dependent decarboxylases (Fenalti et al., 2007) being absent from the electron density (approximately residues 340-350). The overall conformation of the two chains was nearly identical (Fig. 1b).
In the crystal lattice, the protein dimers form layers in the xy plane, separated by a rather uniform spacing (Fig. 1c). As the first $30 amino acids of the tagged protein were not visible, they are most likely to occupy the space between protein dimers in the crystal. This is the likely source of the high mosaicity and incomplete lattice arrangement in the current MmGADL1 crystals.
Both active sites in the dimer are occupied by the PLP cofactor, which is covalently bound to the N " atom of Lys314 in each chain through a Schiff base linkage, being located at the dimer interface (Fig. 1d). Closer observation of the activesite cavity reveals that only the active imine of the linked PLP  is solvent-exposed, and it resides at the end of a narrow cavity, which represents the substrate-binding pocket (Fig. 2a). Electrostatic surface analysis reveals the GADL1 active site to have a high positive charge potential (Fig. 2b). This is logical, as the substrates of GADL1 are acidic amino acids; the basic nature of the binding cavity is involved in electrostatic interactions that attract and bind the substrate, orienting it correctly for catalysis.
In our earlier work, we showed that GADL1 can use aspartate and CSA as substrates, but not glutamate (Winge et al., 2015). While the catalytic cavities of GAD and GADL1 are very similar, it is currently difficult to pinpoint which features of the active site might be responsible for selectivity between such similar substrates. High-resolution structures of GADL1 and its closest homologues with bound active-site ligands will clearly be required. Importantly, a large part of the active-site cavity wall will be formed by the flexible loop in the substrate-bound state; the flexible loop is invisible in most PLP-dependent decarboxylase crystal structures, but harbours a Tyr residue that is likely to be important for catalysis.  Table 3 Small-angle X-ray scattering. 196 (1.6) Molecular-mass determination Molecular mass from I(0) using p(r) (kDa) (ratio to theoretical monomer) 121.0 (2.0) Molecular mass from I(0) using Guinier (kDa) (ratio to theoretical monomer) 120.6 (2.0) Theoretical monomeric molecular mass from sequence (kDa) 60.4 Shape and atomistic modelling CRYSOL (comparison to crystal structure) 2 value versus crystal structure 41.9 s range (nm À1 ) 0.018-4.607 GASBOR (ab initio chain-like modelling) 2 value 1.7 s range (nm À1 ) 0.143-4.607 Symmetry P2 SREFLEX (modelling of flexibility based on crystal structure) 2 value 5.7 s range (nm À1 ) 0.018-4.607 Software Data processing and basic analyses SASFLOW Blanchet et al., 2015) and PRIMUSqt (Petoukhov et al., 2012) Distance distribution analysis GNOM (Svergun, 1992)

MmGADL1 adopts an open conformation in solution
We used SAXS to validate the crystal structure and to determine the conformation of MmGADL1 in solution (Fig. 3, Table 3). As is apparent from the scattering data and the dimensionless Kratky plot, GADL1 presents a folded shape. While the molecular mass calculated from the forward scattering intensity accurately matches that of a dimer, the theoretical scattering curve calculated from the crystal structure differs significantly. The shape in solution is more elongated than in the crystal. The radii of gyration between the theoretical scattering curve from the crystal structure and the experimental value from Guinier analysis differ by nearly 1 nm, indicating a large difference in conformation. The maximum diameter is 3 nm larger in solution than in the crystal state.
The GASBOR chain-like dummy residue model of MmGADL1 is elongated compared with the crystal structure (Fig. 3e). Attempts to model the missing N-terminal portion, while keeping the dimeric crystal structure fixed, did not fit the experimental SAXS data well (data not shown). We thus employed the recently described SREFLEX method (Panjkovich & Svergun, 2016) to take advantage of normal-mode analysis of the crystal structure in SAXS modelling. The SREFLEX model fits the scattering data well and suggests a clearly loosened solution conformation (Fig. 3f ), in contrast to the compact globular structures observed for PLP-dependent decarboxylases in the crystalline state. The open conformation is not similar to the open conformation observed for DOPA decarboxylase in the crystalline state (Giardina et al., 2011;Fig. 3g), in which case the opening occurs in the centre of the dimer. The observed open-close movement is likely to be of functional relevance in the GADL1 catalytic cycle. Whether the different conformations are linked to the binding of ligands remains to be studied. While our GADL1 preparation is enzymatically active (Winge et al., 2015), and the crystal is apparently fully occupied with covalently bound PLP, we cannot currently rule out the presence of PLP-deficient GADL1 in the purified SAXS sample, since crystallization might have enriched a cofactor-bound conformation of the protein.

Comparison to other PLP-dependent decarboxylases
While MmGADL1 and its homologues show low sequence conservation, apart from a few fully conserved core elements  around the active site (Fig. 4a), comparison of the structures of MmGADL1 and its homologues reveals a conserved structural fold with little variance in the arrangement of the PLPlinked Lys residue (Figs. 4b and 4c, Table 4). The sequence conservation between MmGADL1, HsCSAD and HsGADs (Fenalti et al., 2007) is higher than those with human histidine decarboxylase (HDC; Komori et al., 2012) and DOPA decarboxylase (DDC; Giardina et al., 2011;Winge et al., 2015). The latter present similar levels of sequence homology to GADL1 as the bacterial enzymes Sphaerobacter thermophilus PLPdependent decarboxylase (StPDD), Vibrio parahaemolyticus GAD (VpGAD) and the tryptamine-synthesizing enzyme from the gut bacterium Ruminococcus gnavus (RUMGNA_ 01526; Williams et al., 2014). Despite low sequence homology, the R. gnavus enzyme displays high structural similarity to GADL1, suggesting conservation of the fold of PLP-dependent decarboxylases involved in neurotransmitter synthesis. It is interesting to note that the absence of PLP in the active site neither alters the overall tightness of the superposed proteins nor changes the position of the conserved Lys in most structures. In the future, experimental solutionstate methods, such as SAXS, may help to distinguish functionally relevant conformational states from possible crystallographic artifacts. These observations conflict with earlier results described for HsDDC in the beginning of its PLP accumulation-dependent conformational landscape, where a more open conformation was evident in the crystalline state with the active-site Lys residue oriented away from the PLPbinding pocket (Giardina et al., 2011). The substrate specificity and physiological function of GADL1 remain enigmatic. However, the conserved structural details and distinct expression patterns of GADL1 suggest  that it plays a specific physiological role. Variants of GADL1 have been linked to lithium response in bipolar disorder patients (Chen et al., 2014), suggesting a role for GADL1 in lithium pharmacodynamics or kinetics. However, these findings have not been replicated by others, and they have been subject to much controversy (Birnbaum et al., 2014;Cruceanu et al., 2015;Kotambail et al., 2015;Winge et al., 2015;Chen et al., 2016).
Owing to their common mechanistic features, many PLPdependent enzymes are able to catalyze multiple biochemical reactions, making it difficult to define their primary biological function (Percudani & Peracchi, 2003). Of the known GADL1 substrates, Asp appears as the most characteristic substrate for GADL1 (Winge et al., 2015), although the K m is relatively high and the catalytic efficiency is low. Nevertheless, the K m of GADL1 for Asp is in the physiological range, and one could speculate that GADL1 is involved in sensing selected metabolite levels. Relatively low binding affinity is a hallmark of many proteins with signalling roles, which have evolved as sensors of ligand availability; such proteins include, for example, the calcium sensors calmodulin and annexin (Kursula, 2014;Monastyrskaya et al., 2007). Conformational flexibility, as observed here for GADL1 in solution, might have relevance in such a function. Comparison of MmGADL1 with other PLP-dependent decarboxylases. (a) Sequence alignment of MmGADL1 with homologous structures. The conserved PLP-modified lysine is indicated (green triangle). Secondary structure elements and sequence numbering correspond to MmGADL1. (b) Stereo image of a structural superposition of PLP-dependent decarboxylase homologues, viewed towards the active-site cavity. The covalently linked MmGADL1 PLP moiety is depicted as red spheres. The black arrow shows the position of the active-site-covering loop, which is disordered in the MmGADL1 crystal structure. (c) Conservation of PLP conformation in the superposed PLP-dependent decarboxylase structures.

Concluding remarks
The physiological functions and enzymatic properties of GADL1 are subject to further studies. The structure of MmGADL1 and its flexibility in solution, coupled to structural conservation with other PLP-dependent enzymes, point towards functional relevance of these features within the enzyme family. Important future work will concentrate on high-resolution structural details of substrate and inhibitor binding by GADL1, and on comparing these with those of CSAD, GAD and other PLP-dependent decarboxylases. Crucial topics to address will include the fine details of substrate specificity determinants in PLP-dependent decarboxylases, as well as the relevance of the conformational changes observed here to the catalytic cycle of this family of enzymes.