Nicotinamide mononucleotide adenylyltransferase uses its NAD+ substrate-binding site to chaperone phosphorylated Tau

Tau hyper-phosphorylation and deposition into neurofibrillary tangles have been found in brains of patients with Alzheimer’s disease (AD) and other tauopathies. Molecular chaperones are involved in regulating the pathological aggregation of phosphorylated Tau (pTau) and modulating disease progression. Here, we report that nicotinamide mononucleotide adenylyltransferase (NMNAT), a well-known NAD+ synthase, serves as a chaperone of pTau to prevent its amyloid aggregation in vitro as well as mitigate its pathology in a fly tauopathy model. By combining NMR spectroscopy, crystallography, single-molecule and computational approaches, we revealed that NMNAT adopts its enzymatic pocket to specifically bind the phosphorylated sites of pTau, which can be competitively disrupted by the enzymatic substrates of NMNAT. Moreover, we found that NMNAT serves as a co-chaperone of Hsp90 for the specific recognition of pTau over Tau. Our work uncovers a dedicated chaperone of pTau and suggests NMNAT as a key node between NAD+ metabolism and Tau homeostasis in aging and neurodegeneration.


Introduction
Phosphorylated Tau (pTau) is the major component of the neurofibrillary tangles that are commonly found in the brains of patients with Alzheimer's disease (AD) and many other tauopathy-related neurodegenerative diseases (Avila, 2006;Hanger et al., 2009;Hanger et al., 2002;Hanger et al., 1991). Tau is an intrinsically disordered protein with a high abundance in neurons (Hirokawa et al., 1996;Konzack et al., 2007). There are six isoforms of Tau in the human central nervous system due to alternative splicing (Goedert et al., 1989). Under the physiological condition, Tau associates with microtubules and modulates the stability of axonal microtubules (Drechsel et al., 1992). Whereas, phosphorylation of Tau by protein kinases such as microtubule affinity regulating kinase 2 (MARK2), causes the release of Tau from microtubule binding, which leads to hyper-phosphorylation and amyloid aggregation of Tau (Ando et al., 2016;Biernat et al., 1993;Drewes, 2004;Drewes et al., 1997;Gu et al., 2013). The amyloid aggregation of Tau is closely associated with the pathogenesis of AD and other tauopathies (Drewes et al., 1997). Different proteins including chaperones (Hsp90, Hsc70/Hsp70) (Dickey et al., 2007a), proteasome (Dickey et al., 2007b) and protein phosphatase 2A (PP2A) (Gong et al., 1993) were found to play important roles in maintaining Tau homeostasis including preventing abnormal hyper-phosphorylation and aggregation, and facilitating pTau degradation (Hanger et al., 2009).
Nicotinamide mononucleotide adenylyltransferase (NMNAT) was initially identified as an NAD + synthase that catalyzes the reversible conversion of NMN (nicotinamide mononucleotide) to NAD + in the final step of both the de novo biosynthesis and salvage pathways in most organisms across all three kingdoms of life (Magni et al., 1999). NMNAT is indispensable in maintaining neuronal homeostasis (Araki, 2004;Conforti et al., 2009). Familial mutations of NMNAT have been found to cause Leber congenital amaurosis 9 (LCA9) (Chiang et al., 2012;Falk et al., 2012;Koenekoop et al., 2012;Perrault et al., 2012) and retinal degeneration (Beirowski et al., 2008). Moreover, NMNAT is closely related to AD and other tauopathies. The mRNA level of human NMNAT2 (hN2), one of the three isoforms of human NMNATs (Raffaelli et al., 2002), decreases in patients of AD (Ali et al., 2016). Abundant hN2 proteins were detected in the insoluble brain fraction of AD patients, which also contains pTau and Hsp90 (Ali et al., 2016). In addition, NMNAT plays a protective role in different cellular and animal models of AD (Ali et al., 2013;Conforti et al., 2014;Fang et al., 2012;Ocampo et al., 2013). Over-expression of different isoforms of NMNAT can significantly reduce the abnormal aggregation (Zhai et al., 2008) and cytotoxicity of pTau and relieve pTau burden in different models of AD (Rossi et al., 2018) and frontotemporal dementia with parkinsonism linked to chromosome 17 (FTDP-17) Ljungberg et al., 2012).
Intriguingly, in addition to the well-studied NAD + synthase activity, NMNAT has been found to be able to retrieve the activity of luciferase from heat-denatured amorphous aggregation suggesting a chaperone-like activity of NMNAT (Ali et al., 2016;Zhai et al., 2008). However, it remains puzzling how a single domain enzyme, that has no similarity to any known chaperones, fulfills a chaperone-like activity. Moreover, it is confusing whether the protective role of NMNAT in the AD animal models comes from its chaperone-like activity or enzymatic activity, given that NAD + is an essential cofactor in cellular processes such as transcriptional regulation (D'Amours et al., 1998;Shogren-Knaak et al., 2006;Zhang et al., 2009) and oxidative reactions (Lewis et al., 2014).
In this work, we demonstrate the chaperone-like activity of NMNAT against the amyloid aggregation of Tau in vitro and in the fly model. By combining multiple biophysical and computational approaches, we reveal the molecular mechanism of NMNAT as a specific chaperone of pTau. Our work provides the structural basis for how NMNAT manages its dual functions as both an enzyme and a chaperone, as well as how NMNAT specifically recognizes pTau and serves as a co-chaperone of Hsp90 for pTau clearance. Our work suggests an interplay of NAD + metabolism and the progression of Tau pathology in aging and neurodegeneration.

Results
The NMNAT family exhibits a conserved chaperone-like activity in preventing pTau aggregation In the preparation of pTau proteins, we used MARK2 to phosphorylate Tau23 and a truncated construct-K19 ( Figure 1A; Gu et al., 2013). The MARK2 phosphorylation sites on Tau23 and K19 were characterized by 2D 1 H-15 N NMR HSQC spectra ( Figure 1A; Figure 1-figure supplement 1). Consistent with previous reports (Schwalbe et al., 2013;Timm et al., 2003), S262 of repeat region 1 (R1), S324 of repeat region 3 (R3), S352 and S356 of repeat region 4 (R4) were phosphorylated in   Figure 1. NMNATs inhibit pTau amyloid fibril formation and specifically bind to the phosphorylated sites of pTau. (A) Domain compositions of Tau23 and K19. The repeat regions are presented as gray boxes. Phosphorylation sites characterized in this work are marked. (B) Inhibition of NMNATs on the amyloid fibril formation of pTau23 (60 mM) by the ThT fluorescence kinetic assay and TEM microscopy. Human NMNATs: hN1, hN2, and hN3; Drosophila Nmnats: PC and PD; mouse NMNAT3: mN3. A gradient concentration of NMNATs were applied as indicated. The data showned correspond to mean ± s.d., with n = 5. The EM images framed in red mean pTau23 only and those in gray mean 1:1 (mol: mol, pTau23:NMNAT). Scale bar: 500 nm. (C) Overlay of 2D 1 H-15 N HSQC spectra of pK19 alone (100 mM, gray) and pK19 titrated by mN3 (200 mM, red). Signals of pSer residues are enlarged. (D) Residue-specific chemical shift changes of pK19 analyzed based on (c). The domain organization of pK19 is indicated and signals of pSer residues are highlighted in yellow. (E) Residue-specific intensity changes of pK19 (red) and K19 signals (light blue) based on (c) and Source data 1. ThT fluorescence profiles of pTau23 in the absence or presence of various NMNATs at the indicated concentrations ( Figure 1B). Source data 2. Residue-specific chemical shift changes ( Figure 1D) and intensity changes ( Figure 1E) of pK19 titrated by mN3. Source data 3. The association and dissociation response of pK19/K19 to mN3 ( Figure 1F).     both MARK2-treated Tau23 and K19. Besides, pTau23 exhibited two additional phosphorylated sites: S413 and S416.
To comprehensively investigate the activity of NMNAT on pTau aggregation, we prepared different isoforms of NMNAT proteins from different organisms, including three isoforms of human NMNATs (hN1, hN2, and hN3), two isoforms of Drosophila NMNATs (a cytoplasmic isoform PD and a nuclear isoform PC [Ruan et al., 2015]) and mouse NMNAT3 (mN3). We first confirmed that the NMNAT proteins contained normal enzyme activities (Figure 1-figure supplement 2). Then, we conducted the ThT fluorescence kinetic assay and transmission electron microscopy (TEM) to monitor their influences on the amyloid aggregation of pTau. The result showed that different NMNAT isoforms generally exhibited potent chaperone-like activity against the amyloid aggregation of both pTau23 ( Figure 1B The results demonstrate that the chaperone-like activity is highly conserved in the NMNAT family across different organisms, which indicates an important biological role of this activity in the protection of pTau from amyloid aggregation.

Mechanism of the interaction between mN3 and pTau
We next sought to investigate the structural basis of the interaction between NMNAT and pTau. Although hN2 represents the most biological relevant isoform, purified hN2 appeared unstable and prone to aggregate in vitro, which hinders us for further structural characterization. Alternatively, we used mN3, which shares high sequence similarity to hN2 (Figure 1-figure supplement 5) and is more stable. We performed solution NMR spectroscopy and used mN3 to titrate 15 N-labeled pK19. The 2D 1 H-15 N HSQC spectra showed a significant overall signal broadening of pK19, which indicates a strong interaction between mN3 and pK19 ( Figure 1C). In particular, the four phosphorylated Ser (pSer) residues showed large signal attenuations and chemical shift changes upon mN3 titration ( Figure 1D,E; Figure 1-source data 2). Residues adjacent to pSer, including regions around a.a. 250, a.a. 320 and a.a. 350, also exhibited prominent signal attenuations ( Figure 1E; Figure 1-source data 2). Especially, repeat sequence R4 that contains two pSer residues showed the largest signal attenuations with I/I 0 < 0.3 ( Figure 1E; Figure 1-source data 2). In contrast, as we titrated non-phosphorylated K19 at the same ratio, only slight overall signal broadening was observed in the three regions ( Figure 1E; Figure 1-source data 2 and Figure 1-figure supplement 6A). Moreover, pTau23, but not Tau23, showed significant chemical shift changes and intensity attenuations mainly on and around the pSer residues upon mN3 titration (Figure 1-figure supplement 6B). These results indicate that the pSer residues of pTau are the primary binding sites of mN3.
Further, to quantitatively measure the binding affinity between NMNAT and pTau, we conducted BioLayer Interferometry (BLI) analysis that is a label-free technology for measuring biomolecular interactions (Rich and Myszka, 2007). We immobilized mN3 on the biosensor tip and profiled the association and dissociation curves in the presence of either pTau or Tau ( Figure 1F; Figure 1source data 3 and Figure 1-figure supplement 7). As we measured, the binding affinity of mN3 to pTau is about one order of magnitude higher than that to Tau. The dissociation constant (K D ) for mN3 to pK19 is~31 mM and to pTau23 is~9.9 mM. In contrast, the K D for mN3 to K19 is~239 mM and to Tau23 is~58.6 mM ( Figure 1F; Taken together, these results indicate that MARK2-phosphorylation significantly enhances the interaction between NMNAT and pTau through the specific interaction between NMNAT and the phosphorylated residues of pTau.   mN3 utilizes its enzymatic substrate-binding site to bind pTau To identify the binding site of mN3 for pTau, we firstly determined the atomic structure of mN3 at the resolution of 2.0 Å by X-ray crystallography (Supplementary file 1). The structure of mN3 monomer is similar to that of hN3 (Zhang et al., 2003) with an r.m.s.d. value of 0.543 Å between Ca atoms ( Figure 2-figure supplement 1A). The catalytic pocket that synthesizes NAD + from NMN and ATP is highly conserved in the NMNAT family ( The crystal contains two mN3 molecules forming a homo-dimer with a buried surface area of 1,075.8 Å 2 in the asymmetric unit ( Figure 2-figure supplement 1B). Consistently, size exclusion chromatography and multi-angle laser light scattering (SEC-MALS) characterized that mN3 forms a dimer (~64 kDa) in solution ( Figure 2-figure supplement 1C), and as the mN3 concentration decreased, no significant dissociation of the dimer was observed (Figure 2-figure supplement  1D). A similar dimer interface is conserved in Bacillus subtilis NMNAT (BsN) (Olland et al., 2002), hN1 (Zhou et al., 2002), and hN3 (Zhang et al., 2003;Figure 2-figure supplement 1B). Note that BsN also exists as a dimer in solution (Olland et al., 2002). Although hN1 and hN3 exist as tetramer and hexamer in solution, respectively, an equilibrium between dimer and hexamer has been observed (Zhou et al., 2002). These indicate that dimer is a functional unit of NMNAT proteins.
To validate that dimerization is required for the function of mN3, we constructed a double mutation of E198P and L217R (referred to as mutation EL) to disrupt the dimer interface. The SEC-MALS result showed that mN3 EL represents a mixture of dimer and monomer in solution ( These results indicate that the dimerization is important for the stability and enzymatic activity of mN3, but less required for the chaperone-like activity. This difference indicates different mechanisms of the dual activities of mN3. To investigate the mechanism of the chaperone-like activity of mN3, we conducted a cross-linking mass spectrometry (CL-MS) with chemical cross-linker BS 3 to covalently link paired lysine residues in spatial proximity (Ca-Ca distance <24 Å ) as pTau and mN3 interact, and then identified the crosslinked segments by mass spectrometry. The result showed 7 pairs of cross-linked segments between pK19 and mN3 with a confidence score of <10 À7 (Supplementary file 2). Lysine residues, K95, K139, and K206, that are involved in the cross-linking of mN3 with pK19, cluster around the entrance of the enzymatic pocket of mN3 ( Figure 2A). The entrance of the enzymatic pocket features a positively-charged patch mainly composed of residues K55, K56, R205 and K206 for the NMN and ATP binding ( Figure 2B; Figure 2-figure supplement 1A). This result implies that mN3 may utilize the same positively charged binding pocket for both pTau and the enzymatic substrates.
To further validate the pTau binding site on mN3, we constructed a quadruple mutation of K55E, K56E, R205E and K206E (referred to as mutation KKRK) to disrupt the positively charged interface. Differential scanning fluorimetry (DSF) confirmed that the mutations did not impair the overall structural stability of mN3 ( Figure 2-figure supplement 2). To test whether the mutations influence the interaction between mN3 and pTau, we titrated pK19 with the KKRK mutant. The HSQC spectrum showed that the KKRK mutations significantly diminished the affinity of mN3 to the three regions around a.a. 250, a.a. 320 and a.a. 350 that contain pSer residues ( Figure 2C; Figure  Furthermore, the KKRK mutations significantly impaired the chaperone-like activity of mN3 against the amyloid aggregation of both pK19 ( Figure 2D; Figure 2-source data 2) and pTau23 (Figure 2-figure supplement 4A). Note that the disruption of the positively charged patch did not completely eliminate the chaperone-like activity of mN3, indicating that other interactions also contribute to the binding of NMNAT to pTau. Of note, there is a hydrophobic area on the periphery of the positive-charge patch ( Figure 2B), implying that hydrophobic interactions may also contribute to the chaperone-like activity of MNNAT to pTau.
Since our NMR data show that segment 349 RVQ(p)SKIG(p)SLDNI 360 , that is shared by both pK19 and pTau23, is a primary binding segment of mN3, we built a complex structure of mN3 and 349 RVQ

hN3 3 3 3 3 3 3 3 3 3 3 3 mN mN mN mN mN3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
R4 segment Source data 1. Residue-specific intensity changes of pK19 signals titrated by mN3 KKRK mutant ( Figure 2C). Source data 2. ThT fluorescence profiles of pK19 in the absence or presence of mN3 WT and KKRK mutant ( Figure 2D).  (p)SKIG(p)SLDNI 360 by Rosetta modeling ( Figure 2E). The complex structural model showed that the phosphorylated segment is well accommodated in the ATP/NMN-binding pocket of mN3. The phosphate groups of pS352 and pS356 orientate toward the positively charged pocket of mN3 and position in the same binding site of ATP and NMN ( Figure 2E). This structure model explains the observation that the KKRK mutant, which impairs the binding of phosphate groups, specifically abolished the binding of mN3 to segment 349 RVQ(p)SKIG(p)SLDNI 360 of R4 ( Figure 2C; Figure 2source data 1). In addition, the chemical shift perturbations of pS262 of R1 and pS324 of R3 were also abolished as titrated by the KKRK mutant of mN3 ( Figure 2-figure supplement 3), which indicates that these pSer residues may bind to mN3 in a similar manner as those of R4. Primary sequence alignment shows that the key positively charged residues identified for pTau binding are highly conserved in the family of NMNATs from different species (Figure 1-figure supplement 5), which suggests that different NMNAT proteins employ a common and conserved interface for pTau binding. Indeed, mutation of the conserved K57 and R274 residues in hN2 severely impaired its chaperone-like activity against pTau23 aggregation (Figure 2-figure supplement 4B).
Taken together, these results indicate that NMNAT adopts a conserved pocket to bind both the enzymatic substrates and pTau. Thus, the binding of NMNAT to pTau is similar to the specific binding of enzyme and substrate.

Competition between pTau and ATP/NMN for NMNAT binding
We next sought to understand how NMNAT spatially organizes its dual functions with the same pocket in a single domain. We have shown that the KKRK mutations of mN3 diminished the chaperone-like activity of NMNAT since they disrupt the positively charged pocket for the binding of phosphate groups. Next, we tested the influence of the KKRK mutations on the enzymatic activity of mN3 on NAD + synthesis. The result showed that the KKRK mutations also eliminated the enzymatic activity of mN3, which is conceivable due to the inefficient binding of the mutant to the phosphate groups of ATP and NMN ( Figure 3A; Figure 3-source data 1). On the other hand, we mutated H22, a key catalytic residue for NAD + synthesis (Saridakis et al., 2001) that positions deep at the bottom of the substrate-binding pocket (Figure 3-figure supplement 1A). The result showed that the H22A mutation resulted in elimination of the enzymatic activity ( Figure 3A; Figure 3-source data 1), which agrees with the previous study on the enzyme activity of NMNAT (Zhai et al., 2006). However, we found that the H22A mutation showed no influence on the chaperone-like activity of mN3 in inhibiting the amyloid aggregation of pK19 ( Figure 3B; Figure 3-source data 2).
To examine the competition of the two activities, we used the BLI analysis and found that as the concentration of NMN increased, the binding of pK19 to mN3 was remarkably weakened ( Figure 3C; Figure 3-source data 3). The EC50 of NMN as a competitor for mN3's interaction with pTau is 501 mM. Consistently, the ThT assays showed that as the concentrations NMN or ATP decreased, the chaperone-like activity of mN3 on the amyloid aggregation of pK19 dramatically increased ( Figure 3D; Figure 3-source data 4). In contrast, as we reversely added pK19 into the enzymatic reaction of NAD + synthesis, no significant influence was observed ( Figure 3E; Figure 3source data 5).
Taken together, our data indicate that the enzymatic substrates (i.e. NMN and ATP) and the chaperone client pTau of mN3 share the same binding pocket with a partial overlap at the phosphate-binding site, while ATP and NMN are superior to pTau on the mN3 binding.

Nmnat protects pTau-induced synaptopathy in Drosophila
To assess the functional relevance of the direct regulation of NMNAT on the abnormal aggregation of pTau in vivo, we examined the protective capability of Nmnat in Drosophila tauopathy models by overexpressing human wild type (Tau WT ) or pathogenic Tau (Tau R406W ) in the visual system using a photoreceptor-specific driver GMR-GAL4 . The expression pattern can be easily visualized due to the highly organized parallel structure of the compound eye: the R1-R6  The data showed correspond to mean ± s.d., with n = 3. (B) Influence of the H22A mutation on the inhibition of mN3 against the amyloid aggregation of pK19. The molar ratio of pK19 to mN3 is 1:0.2. The data showed correspond to mean ± s.d., with n = 3. Values are compared using Student's t-test. 'ns', not significant. (C) NMN weakens the binding of pK19 to mN3 in a dose-dependent manner measured by BLI analysis. mN3 was immobilized on the SA sensor. pK19 (50 mM) was pre-mixed with NMN at the indicated molar ratios for association. The same amounts of NMN were used for the association and dissociation measurements. The association and dissociation profiles are divided by a vertical dash line. (D) The presence of NMN (top) or ATP (bottom) reduces the inhibitory effect of mN3 against pK19 amyloid aggregation in a dose-dependent manner. Molar ratios of pK19 to NMN/ATP are indicated. The data showed correspond to mean ± s.d., with n = 3. (E) The presence of pK19 (pK19: NMN/ATP = 10:1) shows no significant influence on the enzyme activity of mN3. The data showed correspond to mean ± s.d., with n = 3. Values are compared using Student's t-test. 'ns', not significant. The online version of this article includes the following source data and figure supplement(s) for figure 3: Source data 1. Enzyme activities of mN3 WT and mutants ( Figure 3A). Source data 2. ThT fluorescence profiles of pK19 in the absence or presence of mN3 WT and H22A mutant ( Figure 3B). Source data 3. NMN weakens the binding of pK19 to mN3 ( Figure 3C). Source data 4. The presence of NMN or ATP reduces the inhibitory effect of mN3 against pK19 amyloid aggregation ( Figure 3D). Source data 5. The presence of pK19 shows no significant influence on the enzyme activity of mN3 ( Figure 3E). photoreceptors have their axons traverse the lamina cortex ( Figure 4A,B, magenta box) and make synaptic contacts at the lamina neuropil ( Figure 4A,B, red box), while R7-R8 photoreceptors extend their axons beyond lamina and project to distinct layers in medullar neuropil ( Figure 4A,B, orange box) (Sato et al., 2013). We found that both pTau WT and pTau R406W aggregated in the brain, which could be suppressed by PD overexpression ( Figure 4A; Figure 4-figure supplement 1A, white arrowheads). Compared to Tau WT , Tau R406W exhibited a more severe retinal degeneration in the lamina cortex ( Figure 4A; Figure 4-figure supplement 1A, white asterisks), which was mitigated by PD overexpression. Hyperphosphorylated and aggregated Tau can impair axonal transport, leading to abnormal mitochondrial distribution and clustering (Ebneth et al., 1998). Consistently, we found that overexpression of Tau species, especially Tau R406W , led to remarkable clustering of mitochondria in the lamina cortex, fragmentation of mitochondria in the lamina, and loss of mitochondria at R7-R8 terminals (Figure 4; Figure 4-figure supplement 1A). Overexpression of PD significantly suppressed mitochondrial clustering in the lamina cortex and restored mitochondria localization at R7 and R8 terminals. Synaptic loss has been characterized as one of the main lesions presented in human tauopathies (McGowan et al., 2006). To evaluate the synaptic integrity, we next examined the distribution of Bruchpilot (Brp), an active zone associated cytoskeletal matrix protein. In wildtype lamina, Brp staining of the cross-sections reveals a pattern consists of repetitive cartridge structure, which is formed by R1-R6 photoreceptor terminals (Figure 4-figure supplement 2A). Tau R406W overexpression resulted in synaptic aggregation of hyperphosphorylated Tau (Figure 4-figure supplement 2A, arrowheads) and~50% reduction of the Brp levels within each lamina cartridge compared with that in the wild type flies, suggesting a severe loss of the active zone structures in the presynaptic terminals. Remarkably, the pTau aggregation and the synaptic phenotype can be suppressed by overexpressing PD, as suggested by significantly restored endogenous Brp levels (Figure 4-figure  supplement 2).
It was previously described that Tau increases F-actin aggregation at synaptic terminals, promotes actin polymerization, and reduces synaptic vesicle mobility (Zhou et al., 2017). Consistently, we found an increased level of F-actin at R7 and R8 terminals with Tau WT or Tau R406W expression, which could be suppressed by PD overexpression (Figure 5A and B).
To dissect the protective mechanism of PD, we used a Drosophila line with UAS-PD WR (W98G/ R224A double mutant) insertion that can be used to express an enzyme-inactive PD. The R224 residue in Drosophila is corresponding to the R205 residue in mN3. Our in vitro data show that PD WR has abolished enzymatic activity and modestly decreased chaperone-like activity (Figure 4-figure  supplement 3). In the Drosophila tauopathy model, compared to wild type PD, PD WR shows modestly increased mitoGFP clustering at the lamina cortex, reduced mitoGFP and increased F-actin accumulation at synaptic terminals, indicating that PD WR has reduced protective capacity in vivo. Notably, we did not find a significant difference in pTau level when overexpressing wild type PD and PD WR (Figure 4-figure supplement 1), suggesting that the decreased protective capacity of PD WR is not due to an alteration of pTau clearance capability. To further examine the interaction between PD and pTau in vivo, we co-expressed Tau R406W and PD/PD WR in the third-instar larval salivary gland cells to take advantage of their larger size and better spatial resolution. We found that both wild type PD and PD WR colocalize with pTau ( Figure 4-figure supplement 4), indicating a direct interaction between PD/PD WR and pTau.
Of note, although both Drosophila Nmnat isoforms PC and PD exhibited potent chaperone-like activity against pTau aggregation in vitro (Figure 1-figure supplement 3; Figure 1-figure supplement 4), they exhibit distinct protective capacities in vivo. While PD shows potent protection against tauopathy as evidenced by reduced pTau level, reduced brain apoptosis, and improved locomotor activity, PC has a minimal protective capacity ( Figure 5-figure supplement 1; Figure 5-figure supplement 2). Therefore, the protective capacity of Nmnat in vivo depends on its subcellular localization, as the cytoplasmically localized PD isoform is likely more accessible to the client pTau protein.
Taken together, our data demonstrate that pTau interferes with synaptic functions via (1) impairing mitochondrial dynamics and localization in neurons, (2) disrupting synaptic active zone integrity, and (3) stimulating F-actin accumulation that restricts synaptic vesicle mobility and release. Nmnat specifically binds to pTau and protects against pTau-induced synaptic dysfunction by regulating pTau aggregation, restoring mitochondria and Brp localization at synaptic terminals, and alleviating pathological F-actin accumulation ( Figure 5C).

mN3 mediates the recognition of Hsp90 to pTau
Previous studies showed that hN2 and Hsp90 co-precipitate with pTau in the brains of AD patients, and exhibit a synergistic effect on the attenuation of pTau pathology in cell models (Ali et al., 2016), Here, we used the single-molecule pull-down (SMPull) assay to identify the interplay between  NMNAT, Hsp90, and pTau. SMPull is a powerful tool to quantitatively detect weak and transient interactions between protein complexes. As shown in Figure 6A, His 6 -tagged Hsp90 was coated on the slide, and the binding of pTau23 can be detected by the fluorescence from Alexa-647-labeled pTau23 using the total internal reflection fluorescence (TIRF) microscopy. The result showed that in the absence of mN3, binding of pTau23 to Hsp90 was only at the basal level similar to that of the blank slide (~20 fluorescent spots per imaging area), which indicates that the interaction between them is very weak and transient. However, the addition of mN3 to the Hsp90/pTau23 system significantly increased the fluorescent spots in a dose-dependent manner ( Figure 6B; Figure 6-figure supplement 1A and Figure 6-source data 1). In contrast, the binding of non-phosphorylated Tau23 to Hsp90 is not affected by the addition of mN3 (Figure 6-figure supplement 1A,B; Figure 6-source data 1). Thus, these results indicate that mN3 mediates the binding of pTau23 to Hsp90.
Furthermore, The BLI analysis showed that Hsp90 directly bound to mN3 with a K D value of~1.93 mM ( Figure 6C; Figure 6-source data 2). However, Hsp90 was not able to differentiate pTau23 from Tau23  Source data 1. The average number of fluorescent counts per imaging area detected by SMPull ( Figure 6B and Figure 6-figure supplement 1). Source data 2. BLI measurements of the binding of mN3 ( Figure 6C) or pTau23/Tau23 ( Figure 6D) to Hsp90. Figure supplement 1. mN3 mediates the binding of pTau23, but not Tau23, to Hsp90. Figure 6-source data 2). Taken together, our data indicate that NMNAT acts as a co-chaperone to assist Hsp90 in the recognition of pTau.

Discussion
NMNAT is distinct from canonical molecular chaperones NMNAT proteins have shown a robust neuroprotective activity in various models of neurodegenerative diseases correlated with the decrease of amyloid protein aggregation (Ali et al., 2013;Brazill et al., 2017;Conforti et al., 2014). In this work, we demonstrate that NMNAT functions similar to a molecular chaperone to protect pTau from amyloid aggregation. Our work uncovers the structural basis underlying the binding of NMNAT with pTau and how NMNAT manages its dual enzymatic and chaperone-like functions. As illustrated in Figure 7, as NMNAT binds its enzymatic substrates, i.e. ATP and NMN, the substrates settle deep inside the pocket with defined interactions with NMNAT. As for the binding of pTau, the phosphorylated residues of pTau can specifically dock into the phosphate binding sites of NMNAT, which partially overlaps with the binding of ATP and NMN. Figure 7. Schematics of NMNAT as a key node between pTau homeostasis and NAD + metabolism. NMNAT functions as both an NAD + synthase involved in NAD + metabolism, and a chaperone-like protein assisting the clearance of pathological pTau deposition. During aging, as the levels of ATP and NMN decrease, the chaperone-like function of NMNAT may show up to antagonize pTau aggregation.
It is important to point out that although phosphorylation remarkably increases the binding affinity of mN3 to pTau, without phosphorylation, mN3 still shows weak interactions with the KXGS motifs of Tau ( Figure 1E; Figure 1-source data 2 and Figure 1-figure supplement 6B). As we stated earlier, there is a hydrophobic area on the periphery of the phosphate-binding surface ( Figure 2B), indicating the role of hydrophobic interactions in the NMNAT-client binding. Indeed, a recent study reported an interaction between NMNAT and mutant Huntingtin (Zhu et al., 2019). Huntingtin is uncharged and not known to be associated with hyperphosphorylation. Therefore, NMNAT-client interaction may be versatile. Our study indicates both electrostatic and hydrophobic interactions are involved in the NMNAT-client binding, while NMNAT specifically binds to pTau through its enzymatic pocket.
NMNAT is distinct from canonical chaperones. Canonical chaperones are either ATP-dependent or ATP-independent. In contrast, although NMNAT is able to bind and hydrolyze ATP, its chaperone-like activity is independent of ATP hydrolysis. NMNAT utilizes the same domain, even a shared pocket, for the binding of ATP and pTau, which explains the confusing previous observation that the neuroprotective effect of NMNAT is independent of ATP hydrolysis, yet requires the integrity of the ATP binding site (Ali et al., 2016). In addition, no large conformational change occurs as NMNAT binding to pTau.
In addition to inhibiting pTau aggregation, NMNAT assists Hsp90 in the selection for pTau, implying that NMNAT serves as a co-chaperone of Hsp90 for pTau clearance. Previous studies identified CHIP (carboxyl terminus of the Hsc70-interacting protein) as an important co-chaperone of Hsp90 for pTau removal (Dickey et al., 2007a;Dickey et al., 2007b). Whereas, CHIP-Hsp90 only recognizes Tau phosphorylated at proline-directed Ser/Thr sites, but not KXGS (MARK) sites (Dickey et al., 2007a). Our study shows that NMNAT-Hsp90 can recognize Tau phosphorylated at MARK2-phosphorylated KXGS sites. Therefore, Hsp90 employs different co-chaperones to handle different species of Tau and maintain Tau homeostasis.

NMNAT links NAD + metabolism and tau homeostasis
Our study demonstrates that the defined positively-charged pocket for phosphate binding of NMNAT gives rise to the binding of pTau with a high affinity comparable to that of an enzyme-substrate binding, but distinct from the weak interaction commonly found in chaperone-client binding (Koldewey et al., 2017). The co-existence of the NAD + synthetic and chaperone-like activity sharing a common surface of NMNAT indicates a connection between NAD + metabolism and Tau proteostasis.
NMN is a key intermediate of NAD + . The levels of NMN and NAD + depend on the activity of nicotinamide phosphoribosyltransferase (NAMPT) that converts nicotinamide and PRPP to NMN and is a rate-limiting enzyme in NAD + synthesis (Stein and Imai, 2014). It has been shown that NAMPT and NAD + levels decline during aging (Stein and Imai, 2014;Gomes et al., 2013). Therefore, we may assume that in young and healthy neurons, where the pTau level is low and the NAD + synthesis is normal, NMNAT would mainly function to generate NAD + . In contrast, in old and degenerating neurons, where pTau increases and NAD + synthesis declines, NMNAT would switch its function to inhibit pTau aggregation and assist Hsp90 for pTau clearance (Figure 7).
The cellular processes of NAD + metabolism, and Tau phosphorylation and proteostasis are complex with multiple nodes of regulation. NMNAT emerges as a critical regulator balancing between the NAD + -involved active metabolic state and the amyloid-accumulating proteotoxic stress state. Such regulation would be particularly important to maintain the structural and functional integrity of neurons.

Protein expression and purification
Genes encoding mN1, mN3, Drosophila PC, PD (gift from Dr. Yanshan Fang) and genes encoding hN1, hN3 (purchased from GENEWIZ, Inc Suzhou, China) were amplified and inserted into pET-28a vector with an N-terminal His 6 -tag and a following thrombin cleavage site. Gene encoding hN2 (purchased from Genewiz, Inc) was cloned into pET-32M-3C derived from pET-32a (Novagen). The resulting plasmid encodes a protein with an N-terminal MBP (maltose-binding protein) and a His 6tag followed by an HRV 3C protease recognition site. Mutations of mN3 including KK (K55EK56E), RK (R205EK206E), KKRK (K55EK56ER205EK206E) and H22A were constructed by site-directed mutagenesis using Q5 Site-Directed Mutagenesis Kit (New England Biolabs). All the resulting constructs were verified by DNA sequencing (GENEWIZ, Inc Suzhou, China). NMNATs and variants were over-expressed in E. coli BL21 (DE3) cells. Cells were grown 2 Â YT medium at 37˚C to an OD 600 of 0.8-1. Protein expression was induced by the addition of 0.2 mM isopropyl-b-d-1-thiogalactopyranoside (IPTG) and incubated at 16˚C for 15 hr. Cells were harvested by centrifugation at 4,000 rpm for 20 min and lysed in 50 ml lysis buffer (50 mM Tris-HCl, pH 8.0, 300 mM NaCl, and 2 mM phenylmethanesulfonyl fluoride (PMSF)) by a high-pressure homogenizer (800-1000 bar, 15 min). We next purified the over-expressed proteins by using HisTrap HP (5 ml) and HiLoad 16/600 Superdex 200 columns following the manufacturer's instructions (GE Healthcare). The purified proteins were finally in a buffer of 50 mM Hepes-KOH, pH 8.0, 150 mM KCl, 10 mM MgCl 2 , and 5% glycerol, concentrated, flash frozen in liquid nitrogen, and stored at À80˚C. The purity was assessed by SDS-PAGE. Protein concentration was determined by BCA assay (Thermo Fisher).
Human Tau23/K19 was expressed and purified on the basis of a previously described method (Barghorn et al., 2005). Briefly, Tau23/K19 was purified by a HighTrap HP SP (5 ml) column (GE Healthcare), followed by a Superdex 75 gel filtration column (GE Healthcare). For 15 N-or 15 N/ 13 Clabeled proteins, protein expression was the same as that for unlabeled proteins except that the cells were grown in M9 minimal medium with 15 NH 4 Cl (1 g l À1 ) in the absence or presence of 13 C 6glucose (2 g l À1 ).

Thioflavin T (ThT) fluorescence assay
Amyloid fibril formation of pK19 and pTau23 were monitored using an in situ ThT-binding assay. The ThT kinetics for amyloid fibrils were recorded using a Varioskan Flash Spectral Scanning Multimode Reader (Thermo Fisher Scientific) with sealed 384-microwell plates (Greiner Bio-One). Client proteins were mixed in the absence or presence of NMNATs and variants in indicated molar ratios in a buffer of 50 mM Tris-HCl, 50 mM KCl, 5% glycerol, 0.05% NaN 3 , pH 8.0, respectively. A final concentration of 50 mM ThT was added to each sample. To promote the formation of amyloid fibrils, 5% (v/v) of fibril seeds (the seeds were prepared by sonicating fibrils for 15 s) were added to pK19 and pTau23, respectively. ThT fluorescence was measured in triplicates with shaking at 600 rpm at 37˚C with excitation at 440 nm and emission at 485 nm.
Transmission electron microscopy (TEM) 5 ml of samples were applied to fresh glow-discharged 300-mesh copper carbon grids and stained with 3% v/v uranyl acetate. Specimens were examined by using Tecnai G2 Spirit TEM operated at an accelerating voltage of 120 kV. Images were recorded using a 4K Â 4K charge-coupled device camera (BM-Eagle, FEI Tecnai).

Nuclear magnetic resonance (NMR) spectroscopy
All NMR samples were prepared in an NMR buffer of 25 mM HEPES, 40 mM KCl, 10 mM MgCl 2 , and 10% (v/v) D 2 O at pH 7.0. NMR experiments were collected at 298 K on Bruker Avance 600 and 900 MHz spectrometers. Both spectrometers are equipped with a cryogenic TXI probe. Backbone assignments of K19 and pK19 were accomplished according to the previously published assignments (Eliezer et al., 2005) and validated by the collected 3D HNCACB and CBCACONH spectra, respectively. These experiments were performed using a~1 mM 15N/13C labeled sample. For HSQC titration experiments, each sample (500 ml) was made of 0.1 mM 15N labeled protein (K19/pK19/Tau/ pTau), in the absence or presence of mN3 at a molar ration of 1:2. All NMR spectra were processed using NMRPipe and analyzed using Sparky (Lee et al., 2015) and NMRView (Johnson, 2004).

Biolayer interferometry (BLI) assay
The binding affinity between mN3 and client proteins was inspected by BLI experiments with Forte-Bio Octet RED96 (Pall ForteBio LLC) (Rich and Myszka, 2007). All data were collected at 25˚C with orbital shaking at 1,000 rpm in 96-well black flat-bottom plates (Greiner Bio-One). A total volume of 200 ml was used for each sample and all reagents were diluted in a buffer of 50 mM HEPES, 150 mM KCl, 10 mM MgCl 2 at pH 8.0. Biotinylated mN3 or Hsp90 (20 mg ml À1 ) was loaded onto SA sensors (ForteBio) for 180 s, followed by a 60 s baseline, and then associated with different concentrations of client proteins for 360 s. The association step was followed by a 360 s dissociation step. All data were processed by data analysis software 9.0 (ForteBio). The competition of NMN with pK19 was monitored with the addition of different concentrations of NMN in both association and dissociation solution. For the association step, 50 mM pK19 was pre-mixed with different concentrations of NMN.
To calculate the EC50 of NMN as a competitor for mN3's interaction with pTau, the BLI data were fit to the equation as described (Lian et al., 2013;Xiao et al., 2019): where B represents the response percentage, [I] represents the concentration of NMN used as a competitor, EC50 represents the concentration of NMN that causes a 50% reduction in the BLI response, and n represents the pseudo-Hill coefficient. The EC50 value of the solution competition is 501 mM, n is 0.63, R 2 = 0.994.

mN3 crystallization, data collection and structure determination
Crystals of mN3 were obtained by the hanging drop vapor diffusion method at 18˚C. The condition of 0.04 M citric acid, 0.06 M Bis-Tris propane, pH 6.0-7.5, 20% PEG3350 yielded the diffraction quality crystals after 2 days. Before data collection, crystals were soaked in a cryoprotectant solution consisting of the reservoir solution and 10% (v/v) glycerol and then quickly frozen with liquid nitrogen. Diffraction data of mN3 was collected at the wavelength of 0.9791 Å using an ADSC Quantum 315 r detector at beamline BL17U of Shanghai Synchrotron Radiation Facility (SSRF). Diffraction data for the crystal was collected at 2.00 Å resolutions, as shown in Supplementary file 1. The intensity sets of the mN3 crystal was indexed, integrated and scaled with the HKL2000 package (Otwinowski and Minor, 1997).
The mN3 structure was solved by molecular replacement method using Phaser (McCoy et al., 2007) in the CCP4 crystallographic suite  with the crystal structure of NMN/ NaMN adenylyltransferase (1KQN) as a template. Several cycles of refinement were carried out using Phenix and Coot Vagin et al., 2004) progress in the structural refinement was evaluated by the free R-factor.The mN3 structure belong to the P21 space group with cell dimensions a = 53.7 Å , b = 80.8 Å , c = 64.5 Å .

Size exclusion chromatography and multi-angle laser light scattering (SEC-MALS)
The weight-average molecular weight (Mw) of mN3 was estimated by SEC-MALS that consisted of an SEC column (KD-806 M, Shodex, Tokyo, Japan), a MALS detector (DAWN HELEOS-II,=658 nm, Wyatt Technologies, USA), and a RI detector (Optilab,=658 nm, Wyatt Technologies, USA). 100 ul mN3 (10 mg/ml) in a buffer of 50 mM Hepes-KOH, pH 8.0, 150 mM KCl, 10 mM MgCl 2 , 5% glycerol and 0.05% NaN 3 were loaded to the SEC column with a flow rate of 0.5 ml/min at 25˚C. The resulting data was analyzed using ASTRA VI software (Wyatt Technologies, USA).

Cross-linking mass spectrometry analysis (XL-MS)
Cross-linking experiments were performed as described previously (Zhou et al., 2002). pK19 was incubated with mN3 at 6:1 molar ratio in a buffer containing 50 mM Hepes-KOH, 150 mM KCl at pH 8.0 for 20 min at 4˚C. Cross-linker BS 3 (Thermo Fisher Scientific, 21585) was added at a 1:8 mass ratio and incubated at room temperature for 1 hr. The reaction was quenched with 20 mM ammonium bicarbonate at room temperature for 20 min. Cross-linking products were analyzed by SDS-PAGE to assess the cross-linking efficiency. For MS analysis, proteins were precipitated with acetone; the pellet was resuspended in 8 M urea, 100 mM Tris (pH 8.5) and digested with trypsin at 37˚C overnight. The resulting peptides were analyzed by online nanoflow liquid chromatography tandem mass spectrometry (LC-MS/MS). And the mass spectrometry data were analyzed by pLink (Yang et al., 2012).

Differential scanning fluorimetry (DSF)
Thermal melting experiments were carried out using a QuantStudio 6 and 7 Flex Real-Time PCR Systems (Life) as described previously (Niesen et al., 2007). The buffer is 50 mM HEPES, 150 mM KCl, 10 mM MgCl 2 , and 5% glycerol at pH 8.0. SYPRO Orange (Thermo Fisher) was added as a fluorescence probe at a dilution of 1:1000. 10 mL of protein mixed with SYPRO Orange (Thermo Fisher) solution (1:1000, 50 mM HEPES, 150 mM KCl, 10 mM MgCl 2 , and 5% glycerol at pH 8.0) to a final concentration of 10 mM were assayed in 384-well plates (Life). Excitation and emission filters for SYPRO-Orange dye were 465 nm and 590 nm, respectively. The temperature was increased by 0.9˚C per minute from 25˚C to 96˚C. The inflection point of the transition curve (Tm) is calculated using protein thermal shift software v1.2 (Thermo Fisher).
Modeling of the complex structure of peptide RVQ(p)SKIG(p)SLDNI and mN3 The 12-amino acid peptide 349 RVQ(p)SKIG(p)SLDNI 360 ((p)S: phosphoserine) was docked into mN3 following the Rosetta FlexPepDock protocol (Raveh et al., 2011) in Rosetta software package (Leaver-Fay et al., 2011). Firstly, the 12-mer peptide mimic RVQEKIGELDNI, in which the two phosphoserines were replaced by two glutamates, was docked to the crystal structure of mN3 (PDB: 5Z9R). We performed docking simulations with the restrains of two phosphate binding sites identified in the crystal structure of human cytosolic NMN/NaMN adenylyltransferase (PDB code: 1NUS) (Zhang et al., 2003). The extended 12-mer peptide mimic was initially placed near the putative phosphate binding site. 5000 models were generated by using FlexPepDock protocol to simultaneously fold and dock the peptide over the receptor surface. In this fold-and-dock step, we imposed the distance restraints to confine the glutamate residues of peptide mimic within the phosphate binding sites identified from the crystal structure. The top models with favorable Rosetta energies and satisfied constraints were selected, and the phosphoserines were modeled back by replacing two phosphoserine mimic residues glutamates. The newly modeled structure was further refined by energy minimization to get rid of potential clash and maintain the identified phosphate binding site. After refinement, the top models ranked by Rosetta energies and constraints were selected for visual inspection.

Enzyme activity assay
Enzyme activity of NMNAT was measured in a continuous spectrophotometric coupled assay by monitoring the increase in absorbance of NADH at 340 nm, The reaction process is as follows Balducci et al. (1995): The reaction solution contains 28 mM HEPES buffer (pH 7.4), 11.2 mM MgCl2, 16 nM semicarbazide-HCl, 0.046 mM ethanol, 1.5 mM ATP, and 0.03 mg/ml yeast alcohol dehydrogenase (Sigma, A7011), and NMNAT or variants. The reaction was initiated by adding NMN to a final concentration of 0.625 mM. All measurements were performed at 37˚C. The activity was calculated using the equation below.

Confocal imaging and processing
Fly brains were imaged using an Olympus IX81 confocal microscope coupled with a 60 Â oil immersion objective. Images were processed using FluoView 10-ASW software (Olympus) and analyzed using Fiji/Image J (version 1.52). Statistical analyses were performed using Graphpad Prism (version 7.04).

Western blot analysis
Ten heads of each genotype were homogenized in radioimmunoprecipitation assay (RIPA) buffer (Sigma-Aldrich, St. Louis, MO, USA). Extracted protein samples were mixed with Laemmli sample buffer containing 2% SDS, 10% glycerol, 62.5 mM Tris-HCl (pH 6.8), 0.001% bromophenol blue, and 5% b-mercaptoethanol, and heated at 95˚C for 10 min. Proteins were separated by SDS-polyacrylamide gel electrophoresis and transferred to a nitrocellulose membrane. After blocking at room temperature for 1 hr, the membrane was incubated with primary antibody at 4˚C overnight, followed by the secondary antibody for 1 hr at room temperature. Imaging was performed on an Odyssey Infrared Imaging system (LI-COR Biosciences) and analyzed using Image Studio (v4.0). Primary antibody dilutions were used as follows: anti-cleaved caspase-3 (Asp175, 1:1,000, Cell Signaling) and anti-b-actin (1:10,000, Sigma-Aldrich).

Negative geotaxis assay
Groups of 10 age-matched female flies of each genotype were placed in a vial marked with a line drawn horizontally 8 cm above the bottom surface. Flies were given 45 min to fully recover from CO 2 anesthesia and were gently tapped onto the bottom of the vial and given 10 s to climb. Flies that passed the 8 cm line were counted. The assay was repeated 10 times, and 10 independent groups (n = 10, a total of 100 flies) of each genotype were tested. To eliminate observer-expectancy bias, the assay was carried out with the examiner masked to the group assignment.
All single-molecule assays were performed in the working buffer including 50 mM NaCl, 50 mM Tris, pH 8.0 and 0.1 mM TCEP at room temperature. Single-molecule imaging was conducted in the working buffer containing an oxygen scavenging system consisting of 0.8 mg/ml glucose oxidase, 0.625% glucose, 3 mM Trolox and 0.03 mg/ml catalase to minimize photobleaching. Slides were firstly coated with a mixture of 97% mPEG and 3% biotin-PEG, flow chambers were assembled using strips of double-sided tape and epoxy. Neutravidin and 20 nM biotin-NTA (Biotium) charged with NiCl 2 sequentially flew into the flow chamber and each was incubated for 5 min in the working buffer. The immobilization of Hsp90 (5 nM) was mediated by surface-bound Ni 2+ . Next, 4 nM Tau23/ pTau23 and various concentrations of mN3 were added and incubated with the immobilized Hsp90 for 10 min before data acquisition. An objective type total internal reflection fluorescence (TIRF) microscopy was used to acquire single-molecule data. Alexa647 labeled Tau23 or pTau23 was excited at 647 nm with a narrow band-pass filter (ET680/40 from Chroma Technology). Single-molecule analysis was performed using software smCamera. Mean spot per image (imaging area 2500 mm 2 ) and standard deviation were calculated from 10 different regions.