The structure of human tyrosine hydroxylase reveals the mechanism for feedback inhibition by dopamine

Tyrosine hydroxylase (TH) is a highly regulated enzyme that catalyses the rate-limiting step in the biosynthesis of dopamine (DA) and other catecholamines. Mutations and dysfunction in this enzyme lead to DA deficiency and parkinsonisms of different severity. An understanding of TH deficiency at the level of structure and stability has been lacking to date, as only structures of truncated TH forms have been available. Here, we used cryoEM to determine the first high-resolution structure of full-length human tetrameric TH in the absence (3.4 Å) and presence (3.8 Å) of the end-product and feedback inhibitor DA bound to the active site. We show that upon DA binding, an α -helix (residues 39-59) included within the flexible N-terminal tail of the regulatory domain, is internalized in the active site. The observed structural changes reveal the molecular basis of the inhibitory and stabilizing DA effect, reversible by TH S40-phosphorylation, which are crucial regulatory mechanisms for catecholamine and TH homeostasis.

Tyrosine hydroxylase (TH) is a highly regulated enzyme that catalyses the rate-limiting step in the biosynthesis of dopamine (DA) and other catecholamines. Mutations and dysfunction in this enzyme lead to DA deficiency and parkinsonisms of different severity. An understanding of TH deficiency at the level of structure and stability has been lacking to date, as only structures of truncated TH forms have been available.
Here, we used cryoEM to determine the first high-resolution structure of full-length human tetrameric TH in the absence (3.4 Å) and presence (3.8 Å) of the end-product and feedback inhibitor DA bound to the active site. We show that upon DA binding, an α-helix (residues 39-59) included within the flexible N-terminal tail of the regulatory domain, is internalized in the active site. The observed structural changes reveal the molecular basis of the inhibitory and stabilizing DA effect, reversible by TH S40phosphorylation, which are crucial regulatory mechanisms for catecholamine and TH homeostasis.

INTRODUCTION
Tyrosine hydroxylase (TH; EC. 1.14. 16.2) catalyses the hydroxylation of L-tyrosine to L-3,4-dihydroxyphenylalanine (L-Dopa), the first and rate-limiting step in the synthesis of dopamine (DA), noradrenaline and adrenaline 1 . In the brain these catecholamines (CAs) are essential neuromodulators involved in processes such as motor control, emotion, reward, biorhythms and learning 2 . Mutations in the TH gene are associated with congenital TH deficiency (THD, OMIM #605407), with phenotypes ranging from L-Dopa responsive dystonia (DRD) and infantile parkinsonism to severe, complex encephalopathy with neonatal onset 3,4 . Furthermore, DA insufficiency from selective death of TH-expressing dopaminergic neurons of the substantia nigra pars compacta is associated with Parkinson's disease 5 .
TH belongs to the non-heme iron-and tetrahydrobiopterin (BH4)-dependent aromatic amino acid hydroxylase (AAAH) family, which also includes phenylalanine hydroxylase (PAH) and the tryptophan hydroxylases (TPH1 and 2). Mammalian AAAHs are homotetramers with a three-domain structure ( Supplementary Fig. 1a): an N-terminal regulatory domain (RD) that consists of a structured ACT (aspartate kinasechorismate mutase-TyrA) domain preceded by a less structured N-terminal tail of varying length; a central catalytic domain (CD) that contains the active site iron and binding-sites for substrate and cofactor; and a C-terminal oligomerization domain (OD) responsible for dimerization and tetramerization 6,7 . The full-length structures of rat and human PAH have recently been solved by X-ray crystallography [8][9][10] and cryoEM 10 . For human and rat TH and human TPH, crystal structures encompassing CD+OD are available 11,12 . For rat TH, an NMR structure of the isolated dimeric ACT 13 is also available ( Supplementary Fig. 1b), which allows preparation of composite models of full-length TH based on SAXS data 7,14 . These TH models present a very dynamic structure, notably at the N-terminal tail and the long loops between the RD and CD.
In humans, a single TH gene gives rise to four TH isoforms (TH1-4) by alternative splicing, resulting in variable length of the flexible N-terminus 1 . TH1 is the most studied and the one used in this work ( Supplementary Fig. 2a), and although it has the shortest N-terminus, its RD still includes ~70 residues preceding the folded ACT (residues . ACT domains are typical in many allosteric enzymes, and since TH is not known to be allosterically regulated, it has been suggested that this domain has lost its original function while maintaining its structure throughout evolution 13,15 . The N-terminal tail (residues 1-70) is considered to be largely disordered and variable across species, however residues 40-49 are highly conserved and followed by a poly-alanine segment of variable length (residues 51-59 in humans) ( Supplementary Fig. 2b). This alanine-rich region has been predicted to have high helical propensity 7 .
The activity and stability of TH are regulated through many mechanisms, notably feedback inhibition by CAs and phosphorylation at serine/threonine residues of the N-terminal tail 16,17 , viewed as key to maintain DA homeostasis. TH is phosphorylated on T8, S19, S31 and S40 by several protein kinases with different site specificities 1,17 . The different phosphorylation sites regulate TH through binding to 14-3-3 proteins (S19), cellular localization to the Golgi and synaptic vesicles (S31), and activity (S40) 1,17,18 . Early preparations of TH from the adrenal medulla revealed copurified CAs in the active site, forming a strong bidentate catecholate-Fe(III) complex 19 .
TH inhibition by CAs is competitive with respect to BH4, and they are released from the active site either by incubation with BH4 or by phosphorylation at S40, which is performed by several kinases, including cAMP-dependent protein kinases (PKAs) 17,20,21 . Thus, the strength of feedback inhibition by DA can be modulated by S40-targeting signalling pathways, e.g. enforced by inhibitory auto-receptors that lower PKA activity 22 .
DA acts not only as a feedback inhibitor, but also it stabilizes the enzyme 23 , an effect that is important for maintaining TH levels in vivo, particularly in the axoterminal compartment. The DA stabilizing effect has been demonstrated in vitro 23-25 and in mouse models of DA deficiency 26,27 . In addition to the direct interaction of DA with the active site iron 19 , mutagenesis and truncation studies have revealed the importance of the N-terminal tail on the high-affinity binding of DA and other CAs 28,29 . Still, due to a lack of detailed structural information for full-length TH, a deep understanding of the molecular determinants of the interaction between TH and DA is missing. In this work we have used state-of-the art cryoEM to obtain the structure of full-length human TH at high resolution, both in the absence and presence of its feedback inhibitor DA.
These structures provide us with a better understanding of the inhibitory and stabilizing roles of CA end-products in the regulation of TH activity and TH turnover.

RESULTS
Three-dimensional reconstruction of the substrate-free TH (apo-TH). The high flexibility of the N-terminal RD and of the RD-CD linker has proved to be a major hurdle for structural and functional analysis of the full-length TH; to date there have only been crystal structures of truncated CD+OD domains (PDBs 1TOH and 2XSN) and the NMR structure of part of the RD domain (PDB 2MDA) ( Supplementary Fig.   1b). We set out to optimize conditions for structural determination of the full-length human TH, this time using cryoEM 30 . Tetrameric recombinant human TH was expressed and purified as in 7 ( Supplementary Fig. 3a). When analysed by gel filtration, the elution profile showed a single peak with an elution volume compatible with the 224 kDa of the tetrameric structure ( Supplementary Fig. 3b). This preparation was used for cryoEM optimization using a 200-keV FEI Talos Arctica located at the Centro Nacional de Biotecnología (CNB-CSIC; Madrid), and the best grid was used for data acquisition on a 300-kV Titan Krios at the DLS-eBIC facility (Oxford) (Supplementary Fig. 3c; Data collection parameters are described in Supplementary Table 1 (Fig. 1a), shows a central tetrameric structure formed by the CD and OD domains that is 110 Å long, 86 Å wide and 38 Å high. The four subunits of this central structure have an asymmetric arrangement that could be explained by the tetramer being formed by a dimer of dimers. The two small masses of the RDs have a dimeric structure and a parallelepiped-like shape of 40 x 40 x 22 Å. The two masses were placed on opposite sides of the central part of the TH structure, allowing full access to the active sites of the enzyme. In the 3D reconstruction, many structural features could be assigned to α-helices and loops in the central structure formed by CD and OD domains, in particular the α-helices involved in tetramerization.
In the RD domains, four α-helices were clearly visible, two of them involved in dimer formation.
However, there were areas of lower resolution, so we sought to determine the local resolution of each area of the map using MonoRes 31 ( Supplementary Fig. 4g), which showed a non-isotropic resolution distribution of the map, between 2 Å at the central part of the tetrameric structure and 10 Å at the RDs. An attempt to improve the local resolution of the latter by treating them as single particles did not result in a noticeable increase to resolution (7.1 Å), suggesting that this dimeric structure has intrinsic flexibility ( Supplementary Fig. 5a). The central structure was also subjected to a further refinement by masking out the RDs, and the final resolution reached 3.4 Å ( Supplementary Fig. 5b). Sharpening programs were used to improve the interpretability of the obtained map and to start the model building, as described in the Methods section and in Supplementary Figs. 5c-e. The final atomic model generated ( Fig. 1c; pdb 6ZZU) was very similar to that described in the crystallographic structure of the CD+OD domains (PDBs 1TOH and 2XSN), showing the presence of iron in its active site with the coordinating residues H330, H335 and E375 (Fig. 1d) 11,32 .
Surrounding residues G292, L293, A296, F299, E331, S367 and Y370 form the BH4 binding site, and R315, S323, W371, S394 and D424 form the substrate binding site, where the latter residue is a specificity determinant for L-tyrosine hydroxylation to L-Dopa in TH ( Supplementary Fig. 6) 32,33 . There was also an extra density where one iron-coordinating water molecule could be placed, which was previously observed in the structure of truncated rat TH 11 . In this model, residues 160-170 corresponding to the linker between the RD and CD were also modelled (marked with red arrows in Fig. 1c).
The final pdb file for the apo-TH tetramer included coordinates for residues 78-497 of the four subunits (pdb 7A2G), but not for the 77 N-terminal residues, which are known to be important in TH regulation.
Structural characterisation of TH in the presence of its product, dopamine. As TH plays a pivotal role in DA synthesis and homeostasis, it is important to understand its regulation 2,34 . The RD is essential here as it conveys communication between feedback inhibition by DA and activation by S40 phosphorylation. It was therefore essential to characterise the RD domain structurally at the highest possible resolution and its positioning in the full-length protein 1 . We therefore set out to investigate the structural differences between apo-TH and TH in the presence of DA (TH (DA)).
The formation of stable TH(DA) was carried out as described in the Methods and used for vitrification following the same conditions as for apo-TH. The cryogrids were first analysed in our 200 keV FEI Talos Arctica, and the best one was used to record a total of 4422 movies on a 300 kV Titan Krios at the ESRF Grenoble facility (data collection parameters are described in Supplementary Table 1). The image processing and subsequent 3D reconstruction procedures followed are described in detail in the Methods section and in Supplementary Fig. 7. The final 3D reconstruction yielded a map at 4.1 Å resolution ( Fig. 2a and Supplementary Fig. 7f), although analysis of the local resolution using MonoRes revealed the same differences in resolution described for apo-TH ( Supplementary Fig. 7g).
The volume obtained showed similar structural features and dimensions to those described for apo-TH (compare Figs. 1a and 2a), except for the presence of an extra cylindrical structure, strongly suggestive of the presence of an α-helix protruding from the central mass and contacting the RDs (red arrow in Fig. 2a). These new densities together with the connections with the RD and the CD+OD were masked, and subsequent refinement and sharpening of this part of the enzyme resulted in an increase of the resolution up to 3.8 Å (Fig. 2b). The quality of this new map allowed atomic model building and refinement with COOT and PHENIX. Validation statistics (Supplementary Table 3) showed good correlation values for the attained resolution. In this map, the density of the extra cylindrical volume presented an even clearer helical shape, revealing a new α-helical region in the N-terminus (the first 70 residues) not visible in apo-TH. Earlier secondary structure predictions and modelling by replica exchange molecular dynamics (REMD) have shown the α-helix propensity of the 45-59 alanine-rich region (see sequence in Supplementary Fig. 8a) 7,13 . Thus, we performed secondary structure predictions using PSIPRED 35 and I-TASSER 36 , which actually pointed to a longer region of α-helical propensity (residues 37-59) ( Supplementary Fig.   8a). In accordance with these predictions, a model of the most probable arrangement of the RD in the presence of DA generated using KORP prediction modelling software 37 revealed a 21-residue α-helix that fit perfectly within the corresponding density of the cryoEM map (dark blue α-helices in Fig. 2c). The final model for the TH(DA) tetramer included coordinates for residues 39-497 for the four subunits ( Fig. 2c; pdb 6ZVP), allowing for the first time the modelling of the V60-A70 loop between the α-helix and the ACT domain.
Structural differences between apo-TH and TH(DA). When the two TH structures were compared, no large rearrangements could be observed after DA binding ( Fig. 3a and movie 1). However, there were some interesting changes, all related to the active site. There was a small mass only present in the TH(DA) structure that we assumed was DA (arrow in Fig. 3b). Since the current resolution did not allow accurate fitting of DA, its atomic structure in the PAH(DA) complex (PDB 5PAH) was used to place it in the TH active site, bound to the iron with a bidentate coordination 38 . DA was stabilised by hydrogen bonds with the iron-coordinating residues H330, H335 and E375, by hydrophobic interactions with L41, P326 and Y370, and by an electrostatic interaction of the amine group with D44 (Figs. 3c and 4). L41 and D44 were within the N-terminal α-helix, which also includes the regulatory S40 and appears as a pertinent region for TH regulation by DA. The presence of DA was also accompanied by a rearrangement of the loop between residues C176 and D196 and a shift of the side chain of W371 towards the active-site iron, through interaction with L42 in the incoming α-helix in TH(DA) (Fig.   3c). We evaluated whether the observed closure of the active site affects L-tyrosine binding by alignment of the structure of substrate and BH4-bound PAH-CD (PDB 1KW0) 39 with the TH(DA) structure ( Supplementary Fig. 6). The docked structure revealed a steric clash of DA with the cofactor, but we did not observe clashes between DA and either L-tyrosine or substrate-binding residues ( Supplementary Fig. 6), in agreement with the DA inhibition being competitive only towards BH4 20 .
The major difference found between apo-TH and TH(DA) involves the above mentioned presence of the N-terminal α-helix (residues 39-59) inserted into and blocking the active centre ( Figure 4), running parallel to and establishing contacts with the D360-E375 helix (L41 with Y370, I42 with both T368 and W371, A45 with S367 and R49 with E364). This α-helix also interacts with the V290-R297 loop (A45 with A296, E48 with both S295 and R297), and with the central region of the Q420-S429 loop (I42 with Y422, and E43 and S40 with D424) (Fig. 4). The involvement of several of these residues in DA binding has been revealed in previous mutational and MSmonitored hydrogen/deuterium exchange studies 28,29,40 .
The binding of the N-terminal α-helix (residues 39-59) to the DA active site enhances the inhibition and stability of TH. To determine whether the α-helix observed in the TH(DA) structure is present in apo-TH despite not being visualised due to its great flexibility, or rather is formed upon DA binding, we measured the secondary structure content of the two TH conformations by synchrotron circular dichroism spectroscopy (SRCD) (Fig. 5a). The far-UV spectra were completely overlapping, and calculation of the secondary structure showed no increase in helix content upon DA binding (Supplementary Table 4 Table 4) showed that whereas THΔΝ43 was almost identical to apo-TH and its secondary structure was not affected by DA binding, THΔΝ70 presented a significantly lower α-helical content. These results strengthen the notion that the N-terminal, alanine-rich α-helix is preformed and not in contact with any structured domain of apo-TH, but gets locked into the active centre upon DA binding, which allowed its visualization by cryoEM.
Apo-TH and TH(DA) were further analysed by crosslinking mass spectrometry (XL-MS), which can provide information about nearby regions and possibly help to elucidate whether the presence of DA stabilizes the protein and produces any detectable rearrangement that would account for the presence of the α-helix. Apo-TH and TH(DA) were incubated in the presence of the crosslinker BS3, and subsequently digested and analysed by MS. More than 300 crosslinked peptides were identified for each sample with a false discovery rate (FDR) < 5%, indicating a broad-spectrum coverage. No major differences were found between the two datasets except for two crosslinks which had several hits in apo-TH and just a few or no hits in TH(DA) (Supplementary Fig.   8b). The two crosslinks occurred between the mobile N-terminal region of TH (K12) and two residues of the core structure, the loop connecting the RD and the CD (K162) and an external α-helix that surrounds the active centre (K204). The larger number of crosslinks in apo-TH compared with TH(DA) speaks of a higher flexibility of the former, which is consistent with the presence of an immobile α-helix in the latter that would limit the movement of the entire N-terminal tail (bottom images in Supplementary Fig. 8b), indicative of increased protein stability.
We then analysed a possible effect of DA on the thermal stability of TH and the two deletion mutants by differential scanning calorimetry (DSC) measurement, which provides the temperatures for onset of thermal denaturation (T onset ) and the transition maximum (T max ) ( We also performed activity assays to analyse the inhibitory effect of DA on fulllength TH and the two truncated forms (Fig. 5f,g). The resulting IC 50  Modelling the structural response to S40 phosphorylation. The feedback inhibition of TH by DA is alleviated by PKA phosphorylation of TH at S40, both in vitro and in vivo 1,17,20 . We prepared S40 phosphorylated TH and, as expected, pS40-TH presented increased IC 50 for DA (25-fold higher than unphosphorylated TH) ( Fig. 5f,g). SRCD analysis of pS40-TH revealed a spectrum very similar to that of the unphosphorylated sample and comparable content of secondary structure elements ( Fig. 5b and Supplementary Table 4), indicating that the decreased affinity for DA may arise from a separation of the helix from the active site rather than from disruption of the helical structure. This separation is expected to result in loss of visualization of the helix in the cryoEM structure as in the apo-TH structure and, therefore, to understand the structural causes for the observed effect of this phosphorylation, we modelled the structure of pS40-TH in the presence of DA (pS40-TH(DA)) (Fig. 6a). The added phosphate did not seem a priori to clash with any other residue in the surroundings, though it was close to D44 and D325, which may exert some electrostatic repulsion over the phosphate and induce a measurable displacement of the helix. To obtain more insights on the possible interactions between the phosphate, DA and enzyme, we performed molecular dynamics (MD) simulations on four structures (DA-free, apo forms TH and pS40-TH, in addition to TH(DA) and pS40-TH(DA)), and let it run for 0.5 µs. As shown in Supplementary Fig. 9a, the four systems seemed to equilibrate within the first 200 ns and then remained relatively stable for the rest of the 0.5 µs simulation. The ACT domain (residues 71-165) showed higher average mobility when compared to the CD, as shown by calculated backbone atom fluctuations ( Supplementary Fig. 9b). There were several changes, but we focused our measurements around the N-terminal part of the helix and around iron and DA. After initial equilibration simulations, we monitored some selected distances, averaged for all subunits (Supplementary Table 6). The evolution of interatomic distances suggested a separation of the upper part of the helix from DA in the apo-TH and phosphorylated forms, the latter most probably associated

DISCUSSION
AAAHs have important functions in the synthesis of biogenic monoamines that are essential for many physiological processes, and disruption of these functions (e.g. through mutations) can lead to severe disorders 32,43 . The dynamic behaviour of the RDs, probably closely linked to their important regulatory roles, has until now hindered detailed structural studies. The structure of full-length TH had been elusive until now, herein solved thanks to the latest advances in cryoEM. The structure of human apo-TH (3.4 Å resolution) (Fig. 1) shows a planar, tetrameric core that comprises the CD and OD domains, and two smaller densities on opposite sides and 15 Å apart from the central structure. This great separation explains not only the difficulty of crystallisation, but also in obtaining homogeneous samples for cryoEM, which is reflected in the lower resolution obtained for the RDs (Supplementary Figs. 4g and 5a).
The high resolution reached in the TH core allowed building of its atomic model Similar to what was found in the recent cryoEM PAH structure 10 , there was nonisotropy in the resolution, with the RDs showing less resolution both because of their inherent flexibility and the dynamic linkers that connect them to the central structure.
However, while the resolution in most of the TH core ranges from 2 to 4 Å, it was only about 5 Å in PAH. Whereas tetramers of TH (and also TPHs) form by leucine-zipper interactions, the many polar residues in the PAH OD ( Supplementary Fig. 2a) may allow for different conformations and a dimer-tetramer equilibrium that would limit the possible resolution. There were also differences in the CD that have to do with its arrangement relative to the central OD, with a 12.7° outward tilt of the CD in TH compared with that of PAH ( Supplementary Fig. 10c, left). By comparing the maps for TH (Figs. 1 and 2) and PAH 10  Substrate activation of PAH has been proposed to lead to dimerization of adjacent dimers, reaching a conformation resembling that seen in TH 8,9,45 .
The ingress of α-helix 39-59 into the active site explains the significant thermal stabilization of TH upon DA binding, as the catecholamine establishes several interactions in the CD (the D360-E375 α-helix and the 290-297 and 420-429 loops. D44 was shown to form an electrostatic interaction with the amino group of DA, which also binds to TH through the catechol moiety by a ligand-to-iron(III) charge-transfer transition 19 . The structural stabilization of the N-terminal residues 39-77 in the DAbound state was necessary for their visualization, as they were not observed in the apo-TH state (whose structural model started at the ACT domain). The detailed structural information on the N-terminus provided by the TH(DA) cryoEM structure permitted us to perform MD simulations to complement the structural understanding of the feedback inhibition by DA and its release by S40 phosphorylation.
Physiologically, TH is activated when S31 or S40 are phosphorylated, but only S40 phosphorylation affects CA binding, leaving this site as the main target for signal mediated activation of TH 17 . In our structural model of TH(DA), S40 was placed on the N-terminal side of the α-helix, at the base of the active site. The MD-simulated phosphorylated structures pointed to an electrostatic repulsion from the nearby E325 as a possible mechanism for separation of the helix (Fig. 6a). This offers a structural explanation to the functional result, since release of the α-helix and a lower affinity for DA would result in higher TH activity (Fig. 5g). All these results allowed us to propose a model in which TH is in an active conformation when the N-terminal region (which includes the α-helix) is free and detached from the main structure (Fig. 6b, I). DA binding to the TH active site favours the interaction of the α-helix with the latter (Fig.   6b, II). Phosphorylation of S40 would generate steric repulsion, likely via E325, which subsequently would force the α-helix to egress from the active centre (Fig. 6b, III).
Subsequent release of the DA molecule would result in an active TH (Fig. 6b, IV).
Nevertheless, it is important to note that in the TH(DA) structure the first 38 residues in the N-terminal tail are not visible, and this part of the sequence hosts determinants for the conformational change effected by phosphorylation such as R37 and R38, forming the recognition site for PKA 1,46 .
Our results also highlight the importance of the α-helix in the regulation of neuronal DA homeostasis for unphosphorylated TH, as an IC 50 ≈ 0.5 µM fits well with  (Supplementary Fig. 2a). Despite the low TH sequence identity among different organisms, the helix is highly conserved 13 , at least going back to nematode TH, with only small variations in the poly-alanine region ( Supplementary Fig. 2b) 13  THD is an autosomal recessive parkinsonian disorder caused by mutations in the TH gene, which are registered in the PND database (www.biopku.org/pnddb/). The effect on TH activity, stability and oligomerization has been previously studied for a large number of mutants 3,14 . As seen in Supplementary Fig. 11a, where the mutations are mapped on the subunit structure according to their classification as type A or B 3,4 : the type A mutations occur more frequently in the CD and OD, while the more severe type B appear located around the active site cavity. Analyses based on prokaryotic expression and prediction of mutation-associated destabilization have provided good correlation to phenotype severity for a large number of type A (L-Dopa-responsive dystonia) mutations, but less so for some of the severe type B (non-responsive infantile parkinsonism) mutations 3 . In particular, the large destabilization caused by Type B mutations R297W and T368M and type A I363T were not predicted based on previous truncated structures 3 (www.biopku.org/pnddb/). As DA-mediated TH stabilization determines the steady-state levels of TH and DA in vivo 26,27 , the full-length TH(DA) structure obtained in this work, revealing the interactions of the N-terminal helix with helix 360-375 and loop 290-297 (Fig. 4), improves our understanding of the structurebased pathogenic mechanism in THD. R297 in the 290-297 loop is at the centre of a crucial H-bonding and electrostatic network that hinges with the 360-375 helix, notably with I363, whereas T368 interacts itself with I42 in the N-terminal helix ( Supplementary Fig. 11b), explaining the deleterious conformational loss-of-function of mutations R297W, I363T and T368M.
This work constitutes a significant step forward in the knowledge of the structure, regulation and stabilization of TH through feedback inhibition by DA and its reversal by PKA-mediated S40 phosphorylation. Determining the full-length structure of apo-TH also allowed us to corroborate the differences in oligomeric organization of the RDs in resting AAAHs. The dimeric assembly in TH is consistent with the stable tetrameric structure of the protein, and the positioning of the N-terminal tail (residues 1-70) far from the active site leaves it as a main player in TH regulation. This new structural knowledge enhances the genotype-phenotype link in THD, and paves the way for development of novel stabilizing/chaperoning therapies to address its deleterious neurological manifestations, very much needed by patients with L-Dopa-unresponsive type B.

Cloning, expression and purification of the TH variants. TH with an N-terminal His-
MBP tag containing a TEV protease-site was expressed from the pETMBP1a/TH plasmid and purified as described 7 , with amylose resin affinity chromatography, followed by cutting with TEV protease and isolation of the tetrameric TH by sizeexclusion chromatography in 20 mM Na-Hepes pH 7, 200 mM NaCl. Constructs for THΔΝ43 and THΔΝ70 were derived from the pETMBP1a/TH construct (Genscript) and also expressed and purified as tetramers essentially as TH 7 . For synchrotron circular dichroism (SRCD), the SEC was performed in 50 mM Na-Phosphate pH 7. TH concentration was determined using a Nanodrop UV-Vis spectrophotometer and the theoretical extinction coefficients ε 280 = 47705, 47705 and 40715 M -1 cm -1 for TH, THΔN43 and THΔN70, respectively. Dopamine binding to TH and S40 phosphorylation. Once TH was purified and its functionality assessed, the protein was incubated with iron and dopamine (DA) to obtain the stable Fe(III)-catecholate complex 19,51 . Essentially, ferrous ammonium sulphate (FAS) was dissolved in degassed water, added to TH (1:1 DA:TH subunit) and incubated for 2 min at room temperature (RT) before DA addition (2:1 DA:TH subunit) with subsequent incubation for 3 min before cryoEM grid preparation and/or SRCD spectroscopy assays. TH was phosphorylated on S40 (control without kinase) essentially as described 3 Table 1 Table 1).
Image processing and three-dimensional reconstruction. All programs used for image processing for obtaining the different 3D models are implemented in the Scipion software platform 56 . First, the beam-induced motion of the movies was corrected using the MotionCorr2 software 57 . After movies alignment, the contrast transfer function (CTF) was calculated and corrected by Gctf ( Supplementary Fig. 4a and 7a) 9 . Particles were automatically picked with Xmipp3 -auto-picking software 58 . To save computational time during the first processing steps and to increase the signal to noise ratio, particles were downsampled by a factor of 4, and extracted with Xmipp3 -extract particles 58 . The extracted particles were subjected to a first 2D classification using Relion 2.0 ( Supplementary Fig. 4b and 7b) 59 and the best classes were subjected to several further rounds of 2D classification, allowing a much better detection of bad particles such as aggregates or particles that were very close to each other.
Different de novo initial models (Supplementary Fig. 4c and 7c) were obtained using both EMAN 60 and RANSAC 61 . Another initial model was obtained by low resolution (60 Å) filtering of the atomic structure of the CD of the human TH (PDB 2XSN). In the case of apo-TH, the first rounds of 3D classification were performed using Relion 2.0 without any symmetry imposition and using the different initial models ( Supplementary Fig. 4d). No significant differences were found among the best class obtained from the low-pass filtered atomic structure and the de novo initial model. Particles belonging to that class were subjected to refinement using Relion 2.0 -3D auto-refine software. Since clear symmetric features were observed in this class, we sought to determine whether C2 or D2 symmetry were applicable and could contribute to better define our 3D models. The application of C2 symmetry resulted in different classes showing good arrangement in the CD and OD, but the mass corresponding to the RD was distorted showing artefactual densities. On the other hand, the D2 symmetry maintained the RD size and shape as expected according to its atomic structure (PDB 2MDA) 13 and to the best volume obtained before symmetry imposition ( Supplementary Fig. 4d). In the case of TH(DA), D2 symmetry was directly applied for 3D classification (Supplementary Fig. 7d). The particles selected after refinement (250,712 particles for apo-TH and 125,033 for TH(DA)) were re-extracted at the original pixel size to continue the image processing. Auto-refine with the original particles ( Supplementary Fig. 4e and 7e) rendered a final 3D map at 3.4 Å and 4.1 Å resolution for apo-TH and TH(DA), respectively, as estimated by the Fourier shell correlation (FSC) method, with a cut-off of 0.143 ( Supplementary Fig. 4f and 7f) 59 .
This approach calculates the cross-correlation between two halves of the 3D map in the spatial frequency shells to give the resolution, but it does not contemplate the flexibility of the protein. For each 3D map, local resolution was then calculated using Xmipp3 - Fig. 4g and 7g). The same sets of particles were also subjected to Relion Bayesian particle polishing, however, no improvement in resolution was observed. To improve the low resolution found in the RD, these domains were extracted and processed as single particles to generate a localized reconstruction of the RD 62 ( Supplementary Fig. 5a). The final refined map at 3.4 Å resolution was used for subtracting the region of interest. First, a mask surrounding the selected part of the map (subparticles from now on) was generated and used to extract the subparticles. An initial volume was generated with Relion-reconstruct 59 and used for 3D-classification of the subparticles. The best class was auto-refined to improve the quality of the data and the resolution. The RD final map resolution obtained reached 7.1 Å. This approach was not successful for TH(DA), probably due to the extra density affecting the proper alignment of the subparticles. Another way of improving the different domains in a protein complex is by masking the density of interest. This approach was used to increase the resolution in the CD and OD. Starting from the best auto-refined volume, a mask was generated that excluded the RD. The density inside the mask was subjected to a 3D classification, auto-refine and further postprocessing to obtain a good density map for next model building steps (Supplementary Fig. 5b). The maps were further subjected to sharpening using Relion post-processing 59 , LocScale 63 and LocalDeblur 64 (Supplementary Fig. 4h and 7h). The combination of all of them provided the best interpretability of the 3D reconstructed maps.
Model building, refinement and validation. First, the CD and OD domain structures from the human homotetrameric structure (PDB 2XSN) were fitted rigidly using Chimera 65 . Although a good fit in the cryoEM density map was obtained, a further flexible fitting step with iMODFIT 66 was performed to optimise the arrangement of some flexible segments. The majority of the RD was modelled from the NMR structure of the rat homodimer (PDB 2MDA) with whom the human version shares 82% amino acid identity ( Supplementary Fig. 2b), using the SWISS-MODEL homology-modelling server 67 . The resolution of the RD precluded accurate model building, but was sufficient to rigid-body fit the domain in a position compatible with its connection with the CD.
Unfortunately, no homologous structure was found for the first 70 N-terminal amino acids of the RD. However, an un-modelled density was observed only in the DA-bound structure that could accommodate a long α-helix structure. The secondary structure predictions obtained from I-TASSER 36 and PSIPRED 35 servers agreed in the existence of an approximately 20-residue long α-helix in this region. Based on these observations, a generic 20-residue helix was modelled and fit into the corresponding density with Chimera. Then, we exhaustively scanned all possible orientations and sequence shifts for the first 70 residues of TH within this generic helix. The scanning included all 50 possible sequence shifts within a 20-residue window, as well as translations of the helix axis from -2 to +2 Å in 0.5-Å steps. Finally, the minimum energy of the N-terminal αhelix configuration was selected using KORP (knowledge-based orientational potential) 37 . Finally, the loops connecting the RD with the CD and the N-terminal helix were predicted with the RCD+ server 68 . This online tool efficiently applies the RCD loop-closure algorithm 69 to generate feasible loop conformations that are refined and ranked in PyRosetta 70 . The loop conformations that best accounted for the observed poor electron density completed the TH model.
The final TH models ( Supplementary Fig. 5c,d) were subjected to a double realspace refinement, first manually using COOT 71 and then, an automatic procedure with PHENIX 1.17.1-3660 72 or REFMAC 73 . Implemented in the CCP-EM software platform 74 . The restraints used in the real-space refinement were both standard (bond, angle, planarity, chirality, dihedral, and non-bonded repulsion), and with some additional restraints (Ramachandran plot, C-beta deviations, rotamer, and secondary structure). A local grid search-based fit was included in the refinement strategy to resolve side-chain outliers (rotamers or poor map fitting). Several rounds of real-space refinement were performed until a stable final model was obtained and subsequently validated. For both apo-TH and TH(DA), several equivalent cryoEM maps were used to combine information and refine the tracing of the whole atomic structure. Validation of the final models was done using the phenix_validation_cryoem module in PHENIX.

Molecular dynamics (MD) simulations.
A total of 8 all-atom unbiased MD simulations of tetrameric TH were carried out for 0.5 µs each. Simulations were performed with and without DA coordinated to the active site iron, as well as with and without phosphorylation of S40 (pS40), for a total of 4 individual systems. The atomic tetramer models were based on the cryoEM structure of TH(DA) by inserting DA with the same orientation and iron-oxygen bonding distances as that of human PAH (PDB 5PAH), and by adding a phosphate group to S40. To enhance sampling and statistics, each state was simulated in parallel, differing in their generated random initial velocities. All atomic models were prepared with Amber 18 and the corresponding Amber14SB forcefield 75 . Parameters for DA as well as iron with coordinating residues were prepared with Antechamber 58 and the general Amber forcefield 76 using a semiempirical model. Protonation states of side chains were assigned based on the 3Dstructure using PROPKA at pH 7.0 72 . For each of the simulations, the system was neutralized using a mixture of Cland Na + counter ions, and the protein was solvated in a periodic truncated octahedron box with TIP3 water molecules 77 , providing 16 Å of water between the protein surface and the periodic box edge. The solute was minimized for 10,000 steps, followed by 10,000 steps of minimization of the whole system with restraints on the protein backbone, and finally 20,000 steps of minimization of all atoms. The protein was then heated to 100 K with weak restraints for 20 ps, and to 300 K for 1 ns. Equilibration with constant pressure and temperature (NPT) of the system was performed for a total of 20 ns prior to the production with reduced restraints on the solute. The production runs lasted 500 ns and were performed at a constant volume and energy. All simulations were run with a 2-fs time step, using SHAKE constraints on bonds involving hydrogens. All simulations were run with GPU acceleration 78 on Nvidia RTX 2080Ti cards, producing on average 48 ns of molecular dynamics per day of the systems simulated. The simulations were analyzed using cpptraj 79 . Distances, clustering and fluctuation profiles were shown at the monomer level, averaged over the 4 subunits from 2 simulations. The K-means algorithm was used for clustering and the conformation representing the largest cluster over the last 50 ns was selected to represent each system.

Statistical analysis.
Statistical analyses were performed with Graphpad Prism software version 8. For SRCD and DSC, ≥3 independent samples were measured and the mean ± SD (standard deviation) were calculated. The homogeneity of variances was confirmed by Brown-Forsythe tests and multiple comparisons were made using one-way analysis of variance (ANOVA) followed by a post-hoc HSD Tukey test. Differences in secondary structure content, thermal onset (T onset ) and temperature maximum (T max ) and half-maximal inhibitory concentration (IC 50 ) of the individual TH forms when compared to themselves with/without DA and to full-length TH in the same state were considered significant when p˂0.05.
In activity assays, 3 samples were measured at each concentration of DA. The plotted curves of mean ± SD were fitted to a four-parameter logistic curve by the For all statistically significant differences, the mean difference, 95% confidence interval of difference and effect size were calculated and reported.

CONFLICT OF INTEREST
The authors declare no conflict of interest.