Structures of TorsinA and its disease-mutant complexed with an activator reveal the molecular basis for primary dystonia

The most common cause of early onset primary dystonia, a neuromuscular disease, is a glutamate deletion (ΔE) at position 302/303 of TorsinA, a AAA+ ATPase that resides in the endoplasmic reticulum. While the function of TorsinA remains elusive, the ΔE mutation is known to diminish binding of two TorsinA ATPase activators: lamina-associated protein 1 (LAP1) and its paralog, luminal domain like LAP1 (LULL1). Using a nanobody as a crystallization chaperone, we obtained a 1.4 Å crystal structure of human TorsinA in complex with LULL1. This nanobody likewise stabilized the weakened TorsinAΔE-LULL1 interaction, which enabled us to solve its structure at 1.4 Å also. A comparison of these structures shows, in atomic detail, the subtle differences in activator interactions that separate the healthy from the diseased state. This information may provide a structural platform for drug development, as a small molecule that rescues TorsinAΔE could serve as a cure for primary dystonia. DOI: http://dx.doi.org/10.7554/eLife.17983.001


Introduction
Torsins belong to the AAA+ (ATPases associated with a variety of cellular activities) ATPase family, a functionally diverse group of enzymes, which are fueled by ATP hydrolysis. AAA+ ATPases organize in structurally distinct fashions and interact with various accessory elements to remodel their protein or nucleic acid substrates (Erzberger and Berger, 2006;Wendler et al., 2012;White and Lauring, 2007). Torsins are poorly understood AAA+ proteins with yet elusive functions and unknown substrates (Laudermilch and Schlieker, 2016;Rose et al., 2015). Among the five human torsins (Tor-sinA, TorsinB, Torsin2A, Torsin3A and Torsin4A), neuronally expressed TorsinA carries the most clinical significance since it is at the root of primary dystonia. Primary dystonia is a devastating neuromuscular disease that is predominantly caused by the deletion of glutamate 302 or 303 (DE) in Tor-sinA Ozelius et al., 1997). The etiology of primary dystonia is poorly understood Granata and Warner, 2010), and there is currently no known cure for it.
TorsinA is an unusual AAA+ ATPase, because, unlike any other family member (Erzberger and Berger, 2006;Laudermilch and Schlieker, 2016;Rose et al., 2015;White and Lauring, 2007), it is localized to the endoplasmic reticulum (ER) and the contiguous perinuclear space (PNS), and because it is not self-activated, but instead needs the AAA+-like proteins Lamina-associated protein 1 (LAP1) or Luminal domain like LAP1 (LULL1) to catalyze ATP hydrolysis (Brown et al., 2014;McCullough and Sundquist, 2014;Sosa et al., 2014). LAP1 is a type-II transmembrane protein, which resides at the inner nuclear membrane (INM) through its association with the nuclear lamina . LULL1 is a LAP1 paralog, which localizes to the outer nuclear membrane (ONM) and the continuous ER, with its N-terminal portion protruding into the cytoplasm . The structurally similar luminal domains of LAP1/LULL1 interact with TorsinA, and they provide an arginine finger to the TorsinA active site to facilitate torsin's ATP hydrolysis (Brown et al., 2014;Sosa et al., 2014). Arginine fingers are key structural motifs of AAA + ATPases because they neutralize the transition state during ATP hydrolysis (Wendler et al., 2012). Since torsins lack arginine fingers themselves, this activation mechanism through LAP1/LULL1 is likely critical for their function. As reported by several labs, the disease mutant TorsinA DE is compromised in binding to LAP1/LULL1 Zhao et al., 2013;Zhu et al., 2010). Clearly, this suggests that a probable cause of primary dystonia is the lack of activation of TorsinA. In line with this suggestion, LAP1 deletion shows a similar phenotype to Torsin DE, and contributes to disease pathology (Kim et al., 2010).
To investigate the molecular basis for primary dystonia as a result of the glutamate 302/303 deletion in TorsinA, we took a structural approach. We obtained high-resolution crystal structures of Tor-sinA as well as TorsinADE, each in complex with LULL1, using a nanobody as crystallization chaperone. These structures likely open a pathway toward rational, structure-based drug design against primary dystonia.

Results
TorsinA is a catalytically inactive AAA+ ATPase (Brown et al., 2014;Zhao et al., 2013), notoriously ill-behaved in vitro, primarily due to its limited solubility and stability. We partially overcame these problems by stabilizing an ATP-trapped E171Q mutant of human TorsinA (residues 51-332) by coexpressing it with the luminal activation domain of human LULL1 (residues 233-470). This resulted in a better behaved heterodimeric complex ( Figure 1A), which, however, was still recalcitrant to our crystallization efforts. To facilitate crystallization, we isolated a nanobody (VHH-BS2) from an alpaca immunized with the TorsinA EQ -LULL1 complex. A stable, heterotrimeric complex of TorsinA EQ -eLife digest A group of enzymes known as the AAA+ ATPase family have a wide variety of roles in the cell. They are able to break down a molecule called ATP and use the energy released to change the structure of other 'target' molecules. TorsinA is one such AAA+ ATPase and is found primarily in nerve cells inside two cell compartments called the endoplasmic reticulum and the perinuclear space. There, TorsinA interacts with one of two proteins -called LULL1 and LAP1 -that activate TorsinA. In this respect, TorsinA differs from other members of the AAA+ ATPase family, which can activate themselves without the need for additional proteins.
TorsinA and other enzymes are made up of building blocks called amino acids. Mutant forms of TorsinA that have lost a particular amino acid cause primary dystonia, an incurable neuromuscular disease. This amino acid is needed for TorsinA to interact with LULL1 and LAP1. Previous studies have revealed the 3D structure of LAP1 on its own, but the structure of TorsinA remained unknown.
One way to study the structure of enzymes is to use a technique called X-ray crystallography. The first step in this technique is to make crystals of the protein of interest. However, it has proved difficult to make crystals of TorsinA. Demircioglu et al. have addressed this problem by using X-ray crystallography to investigate the structure of TorsinA when it is bound to LULL1. The experiments used a small molecule known as a nanobody that can specifically recognize the human TorsinA enzyme. The nanobody helped TorsinA to stay attached to LULL1 and form the crystals needed for X-ray crystallography.
The 3D structures reveal how TorsinA and LULL1 interact in a high level of detail, helping to explain how TorsinA differs from other AAA+ ATPases. In addition, by comparing how normal TorsinA and the mutant form interact with LULL1, Demircioglu et al. provide more evidence that primary dystonia is likely to be caused by the improper activation of TorsinA. The subtle differences revealed by these structures could be exploited to develop new drugs to fight this disease in the future.  Figure 1 continued on next page LULL1-VHH-BS2 was readily crystallized in the presence of ATP. We collected a 1.4 Å dataset and solved the structure by molecular replacement, using the LULL1-homolog LAP1 and a VHH template as search models (Sosa et al., 2014) (Materials and methods, Table 1). TorsinA EQ adopts a typical AAA+ ATPase fold ( Figure 1B, Figure 1-figure supplement 1). The N-terminal nucleotide-binding or large domain (residues 55-271) is composed of a central five-stranded, parallel b-sheet surrounded by 8 a-helices. A small three-helix bundle at its C-terminus (residues 272-332), forms critical contacts with LULL1. An ATP molecule is bound in the manner characteristic of P-loop NTPases (Wendler et al., 2012). The Walker A and B motifs are positioned to mediate the requisite nucleotide interactions, with sensor 1 and sensor 2 regions sensing the g-phosphate and thus the nucleotide state ( Figure 1C). The luminal LULL1 activation domain (residues 236-470) adopts an AAA+-like conformation, very similar to its paralog LAP1 (rmsd 1.04 Å over 211 Ca positions, Figure 1-figure supplement 1). The AAA+-like domain comprises a central b-sheet embedded within six a-helices ( Figure 1B). A C-terminal small domain is not present. Similar to LAP1, an intramolecular disulfide bond forms at the C terminus of LULL1, between conserved residues C310 and C468 (Figure 1-figure supplements 1,3). Characteristically, LULL1 lacks nucleotide binding due to the absence of Walker A and B motifs (Sosa et al., 2014). LULL1 forms a composite nucleotide-binding site with TorsinA by providing arginine residue 449 ('arginine finger') at the base of helix a5 ( Figure 1C). The arginine finger activates ATP hydrolysis by TorsinA (Brown et al., 2014;Sosa et al., 2014). The small domain of TorsinA, including helix a7 featuring glutamates 302 and 303, is intimately involved in LULL1 binding. Nanobody VHH-BS2 binds both TorsinA and LULL1 at a shallow groove ( Figure 1B, Figure 1-figure supplement 4). Nanobodies contain three complementarity determining regions (CDRs), with CDR3 most often making critical contacts with the antigen (Muyldermans, 2013). Indeed, the long CDR3 of VHH-BS2 (residues 97-112) is the main binding element in the complex. AAA+ ATPases are organized into a number of structurally defined clades (Erzberger and Berger, 2006;Iyer et al., 2004), distinguished by shared structural elements. Comparison with other AAA+ ATPase structures shows that TorsinA fits best into a clade that also contains the bacterial proteins HslU, ClpA/B, ClpX, and Lon (HCLR clade), all of which are involved in protein degradation or remodeling (Erzberger and Berger, 2006). These AAA+ family members share a b-hairpin insertion that precedes the sensor-I region (Figure 1-figure supplement 1). TorsinA also contains this structural element, but it adopts a distinctly different orientation compared to other members of the clade; however, the pre-sensor I region may be affected by crystal packing in our structure. Two other distinct regions are present. The protein degrading or remodeling AAA+ ATPases all form hexameric rings with a central pore (Hanson and Whiteheart, 2005;Olivares et al., 2016;White and Lauring, 2007). 'Pore loops' in each subunit, conserved elements positioned between strand b2 and helix a2, are critical for threading the protein substrates through the ring (Sauer and Baker, 2011). Torsins are devoid of a pore loop consensus motif (Figure 1-figure supplements 2,5). TorsinA has two cysteines (Cys280, and Cys 319, which is part of the sensor-II motif), positioned near the adenine base of the ATP molecule ( Figure 1D). These cysteines do not form a disulfide bridge in our structure. However, the conservation of Cys280 and the Gly-Cys-Lys sensor-II motif at position 318-320 (  redox activity as part of the ATPase cycle therefore seems highly likely, as has been previously speculated (Zhu et al., 2008(Zhu et al., , 2010. The interaction of TorsinA with its ATPase activators LULL1 and LAP1 is of particular importance, as a prominent mutation causing primary dystonia-the deletion of glutamate 302 or 303-weakens these interactions Zhao et al., 2013;Zhu et al., 2010). But why and how? The TorsinA-LULL1 interface extends over an area of 1439 Å 2 . The main structural elements involved in this interaction are the nucleotide-binding region as well as the small domain of TorsinA, and helices a0, a2, a4 and a5 of LULL1 ( Figure 1, Figure 1-figure supplements 2,3, Figure 2A). The exact position of the small domain of TorsinA relative to the large domain is likely dictated by the sensor II motif, preceding a8, which directly contacts the g-phosphate of ATP through Lys 320, thus serving as an anchor point. A switch to ADP presumably weakens this connection, such that the small domain would become more loosely attached to the large domain. This could explain the observed ATP-dependency of LAP1/LULL1 binding Naismith et al., 2009; Zhao et al., 2013;Zhu et al., 2010). Within the small domain, helix a7, the following loop, and the terminal helix a8 contain all the critical residues. Glutamate 302 and 303 are positioned at the very end of helix a7, and both are involved in TorsinA contacts. Specifically, Glu 303 forms a prominent charge interaction with Arg 276 of LULL1. TorsinA Lys113 -LULL1 Glu385, TorsinA Asp316 -LULL1 Arg419, TorsinA Lys317 -LULL1 Glu415, TorsinA Asp327 -LULL1 Lys283 are additional charge interactions.
To investigate the atomic details of the weakened binding of TorsinADE to LAP1/LULL1, and thus the molecular basis of primary dystonia, we made use of the observation that VHH-BS2 also stabilizes the TorsinA EQ DE(ATP)-LULL1 interaction. We were able to crystallize TorsinA EQ DE(ATP)-LULL1-VHH-BS2 and determine its structure at a resolution of 1.4 Å . Not surprisingly, the overall structure is almost identical to the wild-type protein (0.34 Å rmsd over 276 Ca atoms for TorsinA, 0.27 Å rmsd over 226 Ca atoms for LULL1), except for critical differences in the TorsinA-LULL1 interface ( Figure 2A). The principal difference is that helix a7 is shortened due to the missing Glu 303, with a slight-but significant-restructuring of the loop that follows to establish the connection with helix a8. For future reference, we suggest renaming the DE mutation DE303, rather than DE302/303, since the position of Glu 302 is effectively unchanged. In the dystonia mutant, the TorsinA Glu 303 -LULL1 Arg 276 charge interaction is lost, and the hydrogen-bonding network involving TorsinA Glu 302, Phe 306 and Arg312, as well as LULL1 Arg412 and Glu416 is disrupted (Figure 2A). To determine the importance of different TorsinA residues for LULL1 binding, we performed a co-purification assay ( Figure 2B,C). His-tagged, ATP-trapped TorsinA EQ (residues 51-332) and mutants thereof were recombinantly co-expressed with LULL1 (residues 233-470), but without VHH-BS2, in bacteria. Binding was tested in a co-purification assay using Ni-affinity. The Torsin EQ DE303 mutation abolishes binding in this assay, as expected ( Figure 2B). Since unbound TorsinA EQ is largely insoluble, absence of binding is not registered as an appearance of TorsinA EQ alone, but rather as a lack of eluted protein complex altogether. Eliminating the salt bridge between TorsinA Glu303 and LULL1 Arg276 does not disrupt the TorsinA-LULL1 interaction ( Figure 2B). However, DMet304 and DThr305 both phenocopy DE303 in abolishing LULL1 binding ( Figure 2C). This is in full agreement with published in vivo data using similar mutants (Goodchild and Dauer, 2004). The intricate network of interactions of the a7-a8 loop of TorsinA is crucial for LULL1 binding. Since the DE mutation causes a local change within the small domain of TorsinA rather than protein misfolding, it may be possible to rescue binding by developing a small molecule that resurrects the weakened TorsinADE-LAP1/LULL1 interaction.
Although TorsinADE303 is the most prevalent mutation that causes primary dystonia, it is not the only one (Laudermilch and Schlieker, 2016;Rose et al., 2015). We examined the structural consequence of all known mutations (Figure 2-figure supplement 1, Table 2). Based on our structural data, we strongly predict that most mutations likely cause protein misfolding or they weaken or abolish LAP1/LULL1 binding. Conversely, the two dystonia-mutations found in LAP1 presumably affect torsin interaction. Our structural data, therefore, clearly support the hypothesis that improper torsin activation is the likely cause of primary dystonia (Kim et al., 2010).

Discussion
The biological function of TorsinA remains enigmatic (Granata et al., 2011;Jokhi et al., 2013;Liang et al., 2014;Nery et al., 2008Nery et al., , 2011. Because TorsinA belongs to the AAA+ ATPase superfamily, with specific homology to the bacterial proteins HslU, ClpX, ClpA/B and Lon, it is generally assumed that TorsinA is involved in protein remodeling or protein degradation (Laudermilch and Schlieker, 2016;Rose et al., 2015). However, a substrate of TorsinA has yet to be identified.
The TorsinA structure enables a more thorough comparison to other AAA+ ATPases, particularly with regard to the functionally relevant oligomerization state. After the discovery that LAP1/LULL1 are Arg-finger containing TorsinA activators with a AAA+-like structure, it seemed reasonable to suggest that TorsinA and LAP1/LULL1 likely form heterohexameric rings ((TorsinA-ATP-LAP1/ LULL1) 3 ) in order to function (Brown et al., 2014;Sosa et al., 2014). However, the predominant oligomeric form of recombinant TorsinA-ATP-LAP1/LULL1 complex in vitro and in solution is the heterodimer (Brown et al., 2014;Sosa et al., 2014). In addition, torsin variants have been reported to occur in various oligomeric forms as detected by Blue Native PAGE (BN-PAGE) (Goodchild et al.,  Figure 3). First, we note that the small domain of TorsinA is essential for LAP1/LULL1 binding ( Figure 2C). This is reminiscent of the related HCLR AAA+ clade members where the small domain is known to be critical for hexamerization (Bochtler et al., 2000;Mogk et al., 2003). The importance of the small domain for oligomerization in the context of torsins has also been discussed recently (Rose et al., 2015). Neither LAP1 nor LULL1 harbor a small domain, arguing against formation of a stable heteromeric ring, or, alternatively, suggesting a ring of substantially different architecture. Second, ring formation is important for AAA+ ATPases that thread their protein substrate through a central pore for refolding or for degradation. This central pore is lined with conserved 'pore loops' that are essential for function (White and Lauring, 2007). Neither TorsinA and its homologs, nor LAP1/LULL1 have 'pore loop' equivalents ( Figure 1-figure supplement 5). TorsinA is therefore unlikely to actually employ a peptide threading mechanism that involves a central pore. Third, the surface conservation of LAP1/ LULL1 also argues against a heteromeric ring assembly. Although the catalytic, ATP-containing interface with TorsinA is well-conserved, the presumptive non-catalytic, nucleotide-free interface is not ( Figure 3B). Importantly and in contrast to LAP1/LULL1, the same analysis for TorsinA shows that its 'backside' is conserved. TorsinA may therefore interact in homotypic fashion with TorsinA, with other torsin homologs, or even with an additional, yet unidentified protein. This could mean that the previously observed hexameric assemblies (Goodchild et al., 2015;Jungwirth et al., 2010;Sosa et al., 2014) may only contain one LAP1/LULL1 unit, and multiple torsin units, a property that the employed assays would not have differentiated. It is also possible, that the reported hexameric assemblies reflect a vestigial, yet physiologically irrelevant property, perhaps just of the evolutionary origin of the Torsin-LAP1/LULL1 system. Taking all the existing data into account, it is suggestive that TorsinA may be an exceptional AAA+ ATPase in that it simply acts as a heterodimer, together with LAP1 or LULL1 functioning as an activator. As long as the biological function and the substrate for TorsinA are unclear, however, the physiologically relevant oligomeric state of TorsinA ultimately remains a matter of speculation. Given the unique properties of TorsinA, keeping an open mind about TorsinA assembly into its functional state is called for, as it may well differ more than anticipated from well-studied AAA+ ATPase systems. The observation that the nanobody VHH-BS2 stabilizes the TorsinADE303-LULL1 suggests that it could possibly be used directly as a therapeutic. After all, it could directly rescue TorsinA activity. There are, however, at least two major problems. First, VHH-BS2 only recognizes the TorsinA-(or TorsinADE303-) LULL1 complex, but not the homologous TorsinA-LAP1 complex. The function of LULL1 is still poorly understood, but a knockdown does not generate an NE blebbing phenotype (Goodchild et al., 2015;Turner et al., 2015;Vander Heyden et al., 2009), which is symptomatic for a TorsinA knockout  or a LAP1 knockdown (Kim et al., 2010). Therefore, resurrecting activation of TorsinADE303 via LULL1 is unlikely to ameliorate the dystonia phenotype. Furthermore, the nanobody interaction site on the TorsinA-LULL1 interface is very likely oriented toward the ER membrane, which can be inferred from the relative positions of the membrane anchor of LULL1 and the hydrophobic, likely membrane-proximal N-terminal region of TorsinA. These topological restraints suggest that the nanobody will not bind in vivo, but that it is of significant use for in vitro studies.

Materials and methods
Constructs, protein expression and purification DNA sequences encoding human TorsinA (residues 51-332) and the luminal domain of human LULL1 (residues 233-470) were cloned into a modified ampicillin resistant pETDuet-1 vector (EMD Millipore). TorsinA, N-terminally fused with a human rhinovirus 3C protease cleavable 10xHis-7xArg tag, was inserted into the first multiple cloning site (MCS), whereas the untagged LULL1 was inserted into the second MCS. Mutations on TorsinA and LULL1 were introduced by site-directed mutagenesis. The untagged VHH-BS2 nanobody was cloned into a separate, modified kanamycin resistant pET-30b(+) vector (EMD Biosciences).
To co-express TorsinA (EQ or EQ/DE), LULL1 and VHH-BS2 for crystallization, the E. coli strain LOBSTR(DE3) RIL (Kerafast, Boston MA) (Andersen et al., 2013) was co-transformed with the two constructs described above. Cells were grown at 37˚C in lysogeny broth (LB) medium supplemented with 100 mg ml À1 ampicillin, 25mg ml À1 kanamaycin and 34 mg ml À1 chloramphenicol until an optical density (OD 600 ) of 0.6-0.8 was reached, shifted to 18˚C for 20 min, and induced overnight at 18˚C with 0.2 mM isopropyl b-D-1-thiogalactopyranoside (IPTG). The bacterial cultures were harvested by centrifugation, suspended in lysis buffer (50 mM HEPES/NaOH pH 8.0, 400 mM NaCl, 40 mM imidazole, 10 mM MgCl 2 , and 1 mM ATP) and lysed with a cell disruptor (Constant Systems). The lysate was immediately mixed with 0.1 M phenylmethanesulfonyl fluoride (PMSF) (50 ml per 10 ml lysate) and 250 units of TurboNuclease (Eton Bioscience), and cleared by centrifugation. The soluble fraction was gently mixed with Ni-Sepharose 6 Fast Flow (GE Healthcare) resin for 30 min at 4˚C. After washing with the lysis buffer, bound protein was eluted in elution buffer (10 mM HEPES/NaOH pH 8.0, 150 mM NaCl, 300 mM imidazole, 10 mM MgCl 2 , and 1 mM ATP). The eluted protein complex was immediately purified by size exclusion chromatography on a Superdex S200 column (GE Healthcare) equilibrated in running buffer (10 mM HEPES/NaOH pH 8.0, 150 mM NaCl, 10 mM MgCl 2 , and 0.5 mM ATP). Following the tag removal by 10xHis-7xArg-3C protease, the fusion tags and the protease were separated from the complex by cation-exchange chromatography on a HiTrapS column (GE Healthcare) using a linear NaCl gradient. The flow-through from the cation-exchange chromatography, containing the protein complex, was purified again by size exclusion chromatography on a Superdex S200 column as at the previous step.
For the non-structural analysis of TorsinA and LULL1 variants, the pETDuet-1-based expression plasmid was transformed into LOBSTR(DE3) RIL cells without co-expressing nanobody VHH-BS2. Ni 2+ -affinity purification was performed as described above and bound protein was eluted. Aliquots from the Ni 2+ -eluate and the total lysate were collected and analyzed by SDS-PAGE gel electrophoresis.
Data collection and structure determination X-ray data were collected at NE-CAT beamline 24-ID-C at Argonne National Laboratory. Data reduction was performed with the HKL2000 package (Otwinowski and Minor, 1997), and all subsequent data-processing steps were carried out using programs provided through SBGrid (Morin et al., 2013). The structure of the TorsinA EQ -LULL1-VHH-BS2 complex was solved by molecular replacement (MR) using the Phaser-MR tool from the PHENIX suite (Adams et al., 2010). A three-part MR solution was easily obtained using a sequential search for models of LULL1, VHH-BS2, and TorsinA. The LULL1 model was generated based on the published human LAP1 structure (PDB 4TVS, chain A), using the Sculptor utility of the PHENIX suite (LULL1 241-470 and LAP1 356-583 share 64% sequence identity). The VHH-BS2 model was based on VHH-BS1 (PDB 4TVS, chain a) after removing the complementarity determining regions (CDRs). The poly-Ala model of TorsinA was generated based on E. coli ClpA (PDB 1R6B) using the MODELLER tool of the HHpred server (Sö ding et al., 2005). The asymmetric unit contains one TorsinA EQ -LULL1-VHH-BS2 complex. Iterative model building and refinement steps gradually improved the electron density maps and the model statistics. The stereochemical quality of the final model was validated by Molprobity . TorsinA EQ DE-LULL1-VHH-BS2 crystallized in the same unit cell. Model building was carried starting from a truncated TorsinA EQ -LULL1-VHH-BS2 structure. All manual model building steps were carried out with Coot (Emsley et al., 2010), and phenix.refine was used for iterative refinement. Two alternate conformations of a loop in LULL1 (residues 428-438) were detected in the F o ÀF c difference electron density maps of both structures, and they were partially built. For comparison, the cysteine residues of TorsinA at the catalytic site (residues 280 and 319 in the TorsinA EQ structure) were built in the reduced and the oxidized states, respectively. Building them as oxidized, disulfide-bridged residues consistently produced substantial residual F o ÀF c difference density, which disappeared assuming a reduced state. Statistical parameters of data collection and refinement are all given in Table 1. Structure figures were created in PyMOL (Schrö dinger LLC).

Bioinformatic analysis
Torsin and LAP1/LULL1 sequences were obtained via PSI-BLAST (Altschul et al., 1997) and Backphyre searches (Kelley and Sternberg, 2009). Transmembrane domains were predicted using the HMMTOP tool (Tusnády and Simon, 2001). LAP1/LULL1 proteins were distinguished based on the calculated isoelectric point (pI) of their extra-luminal portions. The intranuclear domain of LAP1 has a characteristically high pI of~8.5-10 due to a clustering of basic residues, while the cytoplasmic domain of LULL1 is distinctively more acidic. Multiple sequence alignments were performed using MUSCLE (Edgar, 2004), and visualized by Jalview (Waterhouse et al., 2009). To illustrate evolutionary conservation on TorsinA and LULL1 surfaces, conservation scores for each residue were calculated using the ConSurf server with default parameters (Glaser et al., 2003).

Generation and selection of nanobodies
The purified human TorsinA EQ -LULL1 complex was injected into a male alpaca (Lama pacos) for immunization. Generation and screening of nanobodies was carried out as previously described (Sosa et al., 2014). Each of the selected nanobodies was subcloned into a pET-30b(+) vector with a C-terminal His 6 -tag. Each nanobody was bacterially expressed and Ni 2+ -affinity purified essentially as described (see above). Different from the TorsinA-containing preparations, MgCl 2 and ATP were eliminated from all buffer solutions. The Ni 2+ -eluate was purified via size exclusion chromatography on a Superdex S75 column (GE Healthcare) in running buffer (10 mM HEPES/NaOH pH 8.0, 150 mM NaCl). Nanobody binding was validated by size exclusion chromatography on a 10/300 Superdex S200 column in 10 mM HEPES/NaOH pH 8.0, 150 mM NaCl, 10 mM MgCl 2 and 0.5 mM ATP. Equimolar amounts of TorsinA EQ -LULL1 and TorsinA EQ -LULL1-VHH were loaded and nanobody binding was monitored by a shift in the elution profile and via SDS-PAGE analysis. After validating VHH-BS2 interaction with TorsinA EQ -LULL1, the C-terminal His 6 -tag of VHH-BS2 was removed from the pET-30b(+) vector for co-purification experiments.

Additional information
Competing interests FED, BAS and TUS: Filed a provisional patent application protecting the use of the crystal structures (U.S.P. T.O. No. 62/330,683 The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Author contributions FED, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article; BAS, Acquisition of data, Analysis and interpretation of data; JI, Acquisition of data, Drafting or revising the article; HLP, Analysis and interpretation of data, Drafting or revising the article; TUS, Conception and design, Analysis and interpretation of data, Drafting or revising the article Author ORCIDs F Esra Demircioglu, http://orcid.org/0000-0002-3866-2742 Thomas U Schwartz, http://orcid.org/0000-0001-8012-1512 Additional files

Major datasets
The following datasets were generated: