NMNAT2 supports vesicular glycolysis via NAD homeostasis to fuel fast axonal transport

Background Bioenergetic maladaptations and axonopathy are often found in the early stages of neurodegeneration. Nicotinamide adenine dinucleotide (NAD), an essential cofactor for energy metabolism, is mainly synthesized by Nicotinamide mononucleotide adenylyl transferase 2 (NMNAT2) in CNS neurons. NMNAT2 mRNA levels are reduced in the brains of Alzheimer’s, Parkinson’s, and Huntington’s disease. Here we addressed whether NMNAT2 is required for axonal health of cortical glutamatergic neurons, whose long-projecting axons are often vulnerable in neurodegenerative conditions. We also tested if NMNAT2 maintains axonal health by ensuring axonal ATP levels for axonal transport, critical for axonal function. Methods We generated mouse and cultured neuron models to determine the impact of NMNAT2 loss from cortical glutamatergic neurons on axonal transport, energetic metabolism, and morphological integrity. In addition, we determined if exogenous NAD supplementation or inhibiting a NAD hydrolase, sterile alpha and TIR motif-containing protein 1 (SARM1), prevented axonal deficits caused by NMNAT2 loss. This study used a combination of genetics, molecular biology, immunohistochemistry, biochemistry, fluorescent time-lapse imaging, live imaging with optical sensors, and anti-sense oligos. Results We provide in vivo evidence that NMNAT2 in glutamatergic neurons is required for axonal survival. Using in vivo and in vitro studies, we demonstrate that NMNAT2 maintains the NAD-redox potential to provide “on-board” ATP via glycolysis to vesicular cargos in distal axons. Exogenous NAD+ supplementation to NMNAT2 KO neurons restores glycolysis and resumes fast axonal transport. Finally, we demonstrate both in vitro and in vivo that reducing the activity of SARM1, an NAD degradation enzyme, can reduce axonal transport deficits and suppress axon degeneration in NMNAT2 KO neurons. Conclusion NMNAT2 ensures axonal health by maintaining NAD redox potential in distal axons to ensure efficient vesicular glycolysis required for fast axonal transport.


Genotyping
For mice, ear lysates were prepared by immersing the tissue derived from a small ear clip in digestion buffer (50 mM KCl, 10 mM Tris-HCl, 0.1% Triton X-100, 0.1 mg/ml proteinase K, pH 9.0), vortexing gently, and then incubating for 3 h at 60°C to lyse the cells. For embryos, the same digestion buffer with 0.2 mg/ml proteinase K was used, and embryonic tail lysates were incubated at 60°C for 15 min with 1500 rpm shaking in a thermoshaker. These lysates were then heated to 95°C for 10 min to denature the proteinase K (Thermo Scienti c, Rockford, IL, USA) and centrifuged at 16,100 g for 15 min. The supernatants were used as DNA templates for polymerase chain reactions (PCRs, EconoTaq Plus Green 2X mater mix, Lucigen, Middleton, WI, USA). For embryonic genotyping during primary culture preparation, a QIAGEN Fast Cycling PCR Kit was used. The sequences of the primers used for genotyping are listed in the reagent table. For NMNAT2 f/f line: primers A and B detect WT allele; primers B and C detect NMNAT f allele. For NEX-Cre line: primer Cre484 and Cre834 detect Cre positive allele. For the NMNAT2-Blad line: primers A and B detect WT allele, and primers R3 and Rf detect KO allele. For SARM1 KO line: primers WT-R and Sarm1-common detect WT allele, and primers Sarm1-common and Mut-R detect KO allele.
Immunohistochemistry for brain sections Mice were anesthetized and perfused with 4% paraformaldehyde (PFA) in PBS. Brains were harvested, post-xed with 4% PFA in PBS overnight at 4°C, and then rinsed with PBS. Free-oating brain sections were prepared by sectioning these xed brains into 40-µm-thick sections in the coronal plane with a Leica VT-1000 Vibrating microtome or Sliding Microtome SM-2000R (Leica Microsystems).
For immunohistochemistry, sections were permeabilized with 0.3% Triton X-100 in PBS for 20 min at room temperature, incubated with blocking solution (3% normal goat serum prepared in PBS with 0.3% Triton X-100) for 1 h, and then incubated with primary antibodies diluted in blocking solution overnight at 4°C. The next day, sections were washed with 0.3% Triton PBS 3 times and then incubated with the secondary antibodies diluted in blocking solution at 4°C for 2 h. After the incubation, samples were washed with 0.3% Triton PBS 3 times. Draq5 (1:10,000 dilution, Cell Signaling) or 4´,6-diamidino-2phenylindole (DAPI, 5 µg/ml, Invitrogen) were added during the rst wash step to visualize nuclei. Dako mounting medium was used to mount the brain sections.
Microscopy, imaging, and data analysis for brain sections.
Bright-eld images were obtained with Zeiss SteREO Discovery.V8 Microscope (Carl Zeiss Microscopy). Confocal uorescent images were taken by a Leica TCS SPE confocal microscope (Leica DM 2500) with a 10x objective lens (0.3 N.A.) or a 40x (0.75 N.A.) objective lens. Some confocal images were taken with a Nikon A1R laser scanning confocal (Nikon A1) with a 10x (0.5 N.A.) or a 60x (1.4 N.A.) oil objective lens. DAPI/Draq5 immuno uorescence was used to identify comparable anatomical regions across different brain sections. A minimum of three sections were imaged per mouse, and each anatomical region was imaged with both sides of the cortices, comparable across all animals. Images for particular staining were acquired with identical imaging parameters to minimize signal saturation in all experimental groups. The thickness of the corpus callosum was measured from the dorsal to ventral edges of NFM-positive axonal tracts in coronal plane brain sections. The thickness of the primary somatosensory cortex was measured from the pial surface to white matter regions using ImageJ "Straight" function. The total pixel values of APP signals and area sizes were measured by Image J to acquire APP signal densities.

NMNAT2 in situ hybridization
In situ hybridization on embryonic sections was done by the Baylor College of Medicine Advanced Technology Core Labs. The experiment was performed using a high-throughput automated platform.
Cryostat sections of E14.5 embryos were placed on a standard microscope slide that was subsequently incorporated into a ow-through chamber. The chamber was then placed into a temperature-controlled rack, and the required solutions for pre-hybridization, hybridization, and signal detection reactions were added to the ow-through chamber with an automated solvent delivery system. Details can be found in a previous publication [5]. The RNA probe was generated using the Allen Brain Atlas database sequence.
Images were acquired by a Leica CCD camera with a motorized stage. Multiple images were collected from the same section. Individual images were stitched together to produce a mosaic representing the entire section.
Immunocytochemistry and confocal imaging About 2.5*10 5 cortical neurons were plated on 12 mm diameter coverslips coated with poly-D-lysine (PDL) in 24-well plates (1.3*10 3 cells/mm 2 ). At indicated days in vitro, cortical neurons were rst xed by 4% PFA and 4% sucrose in PBS for 20 min, incubated for 60 min in blocking buffer (0.1% Triton X-100 and 5% Goat Serum in PBS), and then incubated with primary antibody containing blocking buffer overnight at 4°C on a gentle rotating platform. Samples were washed with 0.1% Triton in PBS 3 times and incubated with a secondary antibody containing blocking buffer for 2 h. After the incubation, samples were washed with 0.1% Triton in PBS 3 times. Coverslips were removed to mount with ProLong™ Gold antifade mounting medium with DAPI.
Fluorescent confocal images were taken by a Leica TCS SPE confocal microscope using a 63x (1.4 N.A.) oil objective lens or by a Nikon A1R-HD25 laser scanning confocal microscope with an Apo Lambda S 60x (1.4 N.A.) oil objective lens. Z-stack images were taken with 0.5 µm Z-step size to cover the whole depth of the cultured neurons. Images were acquired with identical imaging parameters chosen to optimize the signal-to-noise ratio and avoid saturated pixels in experimental groups. To quantify APP or TUJ1, 3-5 regions were randomly selected per coverslip for imaging. Each image accounts for one data point in quanti cation. Two coverslips per group were imaged for each batch of culture. To sample a larger area for imaging MAP2 and βIII-tubulin staining for gross neurite area quanti cation, the "large image stitching" function in Nikon NIS-Elements software with 5-10% overlap was used to tile 4 adjacent elds of view together as one image.
Quanti cation of APP accumulation and gross neurite area in neuronal culture ImageJ was used for quanti cation of the maximum intensity projected images. APP signal in somata and dendrites was manually excluded, then the default threshold method was applied to select the top 10% tail of total pixels, and the "analyze particles" function was applied to select particles of area larger than 0.9 µm 2 which were identi ed as "accumulated APP". The total area of "accumulated APP" was calculated in each image to evaluate the phenotype. MAP2-positive areas were identi ed using an autothresholding method and measured for each image. A MAP2 positive area was selected by the "create selection" function after thresholding, then restored and deleted from the βIII-tubulin channel. Therefore, the remaining βIII-tubulin signal solely represented axonal regions. Similarly, mean threshold was applied to measure the βIII-tubulin positive area. Live cells were distinguished from dead cells by DAPI-stained nuclei morphology, and the number of live cells was counted for each image. MAP2 and βIII-tubulin areas were normalized by dividing by the live cell number in each image. We observed a skew in the data if the cell density dramatically differed between Ctrl and KO. Therefore, only images with a live cell number between 150 to 300 within 0.31 mm 2 image area (485-970 cells/mm 2 ) were used for the analysis.

Transfection in neuronal culture
For lentiviral transduction in NMNAT2 f/f neurons, a lentivirus expressing copGFP or iCre under elongation factor 1 alpha (ELF1alpha) promoter was applied at 2 multiplicity of infection (MOI) at 9 days in vitro (DIV9).
For lipofectamine transfection, about 1.6*10 6 cortical neurons were plated onto PDL (Poly-D-Lysine)- Kymographs were generated using ImageJ (http://rsb.info.nih.gov/ij) with "Velocity_Measurement_Tool" macro (http://dev.mri.cnrs.fr/projects/imagej-macros/wiki/Velocity_Measurement_Tool). Velocity quanti cation was manually performed by following the trajectory of each particle with an angle larger than 0° relative to the time axis. Stationary and repetitive bidirectional-moving particles were excluded for velocity measurements. If an anterograde or retrograde moving particle stopped brie y during the imaging and then continued moving, the paused portion was included in the velocity calculation. Although a portion of the axon could be out of focus, resulting in a discontinuous trajectory, the discontinuous trajectory was still traced as the same if the slope and timing matched. All the traced trajectories were saved in "ROI manager" in ImageJ. Particles that remained stationary or underwent repetitive bidirectional movements for most of the recording time were de ned as stationary/dynamic pause events. Each trajectory of stational/dynamic pause particles that could be visually separated was included in the number of stationary/dynamic pause events. The number of anterograde or retrograde events was counted in the same way. The stationary/dynamic pause percentage was calculated as the number of stationary/dynamic events divided by the sum of the number of anterograde, retrograde, and stationary/dynamics pause events. Anterograde and retrograde percentages were calculated similarly.

Colocalization analysis
Control neurons were plated at a density of 1.8*10 3 cells/mm 2 on PDL-coated 12mm diameter coverslip in 24-well culture plate (about 3.5*10 5 cells on 190 mm 2 surface). Through transfection, pmCherry-n1-NMNAT2 was co-expressed with pEGFP-n1-APP, pEGFP-n1-SYPH, and pEGFP-c1-SNAP25 at DIV6, and samples were xed using 4% PFA with 4% sucrose dissolved in PBS at DIV8. Immunocytochemistry was conducted to amplify the mCherry and EGFP signals after xation. ProLong™ Gold antifade mounting medium with DAPI was used to mount the coverslips. Fluorescent confocal images were taken by a Leica TCS SPE confocal microscope using a 63x (1.4 N.A.) oil objective lens with a system-optimized z-step size. The colocalization was further validated by structure illumination imaging using OMX-SR 3D-SIM Super-Resolution System with a 60x (1.516 N.A.) oil objective lens, following system-optimized z-step size.
JACoP (Just Another Colocalization) ImageJ plugin was used to analyze confocal images. Brie y, cell debris signal outside axon was manually deleted. "Objects based methods" was used. The threshold for both channels was manually adjusted based on the visual judgement of signal coverage. The minimum particle size was set as 25 pixels. The ratio of particles colocalized to NMNAT2 was calculated as the number of colocalizing particles in APP, SYPH, or SNAP25 channels divided by the total number of particles in these channels.

Mitochondria density and morphology analysis
Both control and KO neurons were plated at a density of 1.8*10 3 cells/mm 2 on PDL-coated 12mm diameter coverslip in 24-well culture plate (about 3.5*10 5 cells on 190 mm 2 surface). Neurons were cotransfected with pCMV-MitoVenus and pCAG-mCherry at DIV6 and xed using 4% PFA with 4% sucrose at DIV8. Immunocytochemistry with antibodies against GFP (Chicken) and RFP (Rabbit) was conducted to immuno-amplify the MitoVenus and mCherry signals after xation. ProLong™ Gold antifade mounting medium with DAPI was used to mount the coverslips.
Images were taken by a Nikon A1 laser scanning confocal microscope using a 1.4 N.A. Apo Lambda S 60x oil objective lens with 0.171 µm z-step size and 3 times zoom to cover the whole depth of the axon segment of interest. Axonal segments at least 400 µm away from the soma were identi ed in the RFP channel and selected for imaging. Distal axons from at least 8-10 neurons were randomly sampled per coverslip and two coverslips were imaged per group. Images were analyzed using Imaris (Oxford Instruments). RFP channel was used to generate the surface masking of the axonal region, within which the MitoVenus signal was used to create the surface representing mitochondrial morphology. To calculate mitochondria density, length of the axon segment in each image was measured. Density was calculated from mitochondria number divided by axonal length in each image. Sphericity values of mitochondrial surface from axon segments of each individual neuron were input in each column in the column table in Prism 9.0 (each column representing one individual neuron). Cumulative frequency distribution of relative percentage was generated through "column analysis" function in Prism 9.0. Statistical analysis of cumulative frequency distribution is described in "quanti cation and statistical analysis" section. NAD + /NADH live imaging using the genetically encoded sensor, SoNar Around 1.66*10 3 cells/mm 2 cortical neurons were plated on PDL (Poly-D-Lysine)-coated MaTek dish (P35G-1.5-20-C). The NAD + /NADH sensor, SoNar, and its control cpYFP were gifts from Yi Yang's group. On DIV6, 2.5 ug DNA of pCDNA3.1-SoNar or pCDNA3.1-cpYFP were transfected. After 48 h expression of the sensor or control (DIV8), the culture medium was replaced by Hibernate E buffer and maintained at 37°C for imaging using a Leica TCS SP8 confocal laser scanning platform with HyD hybrid detector and an HC PL APO 40x (1.2 NA) water-immersion objective. The SoNar sensor/cpYFP was excited at 440 nm and 488 nm, and the emission at 530+/-15 nm was detected to obtain the ratiometric measurement. Laser power intensity was maintained the same between WT/Het and KO in the same experiment. Axonal segment > 400 µm away from the soma (de ned as distal axon) and axonal segment within 200 µm from the soma (de ned as proximal axon) were imaged with 1 µm z-step size to cover the whole axon segment within the eld of view. Maximum intensity projection was generated for further analysis. Mean grey value from 488 nm excitation (F488) and 440 nm excitation (F440) was measured in ImageJ using manually drawn regions of interest (ROIs) that circle the edges of the axons of interest. Five rectangular ROIs were randomly drawn in the background area and the average mean grey value from these ve ROIs was used to subtract background. After background subtraction, the ratio was calculated as (F488 axon -F488 background )/ (F440 axon -F440 background ). Data presented were normalized to control as a percentage.

Syn-ATP live imaging
Syn-ATP imaging was performed using a custom-built inverted spinning disk confocal microscope (3i imaging systems; model CSU-W1) with two cameras: an Andor iXon Life 888 for confocal uorescence imaging and an Andor iXon Ultra 888 electron-multiplying charge-coupled device camera for luminescence imaging. The Andor iXon Ultra 888 camera was selected for ultralow dark noise, further reduced by cooling to − 100°C. The Andor iXon Ultra 888 camera speed used for luminescence measurements was 1 MHz with 1.00 gain and 1000 intensi cation. Image acquisition was controlled by Slidebook 6 software. Confocal imaging of mCherry uorescence was performed by illuminating neurons with a 561 nm laser with 200 ms exposure and 2.3 mW laser power. For mCherry uorescence measurements, ten frames of timelapse imaging without interval were acquired and the average of ten images was used to measure the mCherry uorescence signal from the varicosities of interest. For luminescence measurements, luminescence photons were collected by accumulating the image for 60 s in the presence of 2 mM D-luciferin (Promega), and the luminescence signal was measured from the same varicosities as the corresponding uorescence image. All images were acquired through a Plan-Apochromat 63x/1.4 N.A. Oil objective, M27 with DIC III prism, using a CSU-W1 Dichroic for 561 nm excitation with Quad emitter and individual emitters for confocal uorescence, and a 720 nm multiphoton short-pass emission lter was used for luminescence. During imaging, the temperature was maintained at 37°C using an Okolab stage top incubator with temperature control. Distal axons at least 400 µm from the soma were selected for imaging.
Images were analyzed in ImageJ using the plugin Time series analyzer luminescence (L background ) channels were used to subtract the background for each varicosity from the uorescence (F varicosity ) and luminescence (L varicosity ) channels, respectively.
The average L/F value across varicosities within one neuron was used to represent that neuron's relative presynaptic ATP level, accounting for one data point in the statistical analysis.
pH measurement and pH correction of L/F measured from Syn-ATP Cytoplasmic pHluorin (Cyto-pHluorin) was transfected in cortical neurons at DIV7, and imaging was done at DIV8 using the same equipment setting as Syn-ATP measurements. Detailed procedures are described in the previous publication [33]. In brief, neuronal culture medium was replaced by Tyrodes buffer containing (in mM) 119  Cytosolic pH of the basal and Oligomycin treatments were determined using the following modi ed Henderson-Hasselbalch equation: Here, the average pH in each condition was used for correction and we did not propagate errors in pH measurements into the nal error shown in L/F measurements. pH corrections of L/F values were only done for those conditions that showed a statistically signi cant change in pH compared to control (as in NAD + supplementation in neuronal culture NAD + (Roche, NAD100-RO) was dissolved in PBS at a stock concentration of 100 mM, sterilized, aliquoted, and stored at -80°C. Only less than 1-week-old stocks were used. On the starting day of supplementation, NAD + was applied to the culture medium at a nal concentration of 1 mM. In the following days before live imaging, 1/3 of culture medium was replaced with a fresh medium containing 1 mM NAD + daily. 1 mM NAD + was supplemented to the Hibernate E low uorescence buffer used for live imaging. In the following days before xation, 1/3 of culture medium was replaced with fresh medium containing 1 mM NAD + every 3 days.

TUJ1 fragmentation quanti cation
Fluorescent confocal images of TUJ1 staining were taken as described in the "immunocytochemistry and confocal imaging" section. TUJ1 fragmentation was analyzed following the degeneration index measurements method described by Kraemer et al.
[48], with some modi cations to their particle size and circularity indexes. TUJ1 fragmentation quanti cation was performed using ImageJ. The uorescent channel corresponding to TUJ1 was isolated and converted to a binary image using the "Make Binary" function, and this image was used for all subsequent steps of the analysis. The total TUJ1 expression area was measured using the "Measure" function. Images were then processed using the "Analyze Particles" function, setting the size parameter at 0.327 µm 2 -817.587 µm 2 (4-10000 pixels) and circularity at 0.2-1. The area of selected particles was recorded and normalized to total area of TUJ1. To analyze cumulative frequency distribution curve, the dependent variable P (the relative percentage within each bin generated by GraphPad Prism "Column analysis" ) was transformed into log(P/(100-P)) so that the relationship between covariate (bin of sphericity) and the transformed dependent variable was linearized. Then, a linear mixed model with a random slope and intercept was generated by SPSS using a restricted maximum likelihood approach to compare statistical difference between control and KO. The criterion for statistical signi cance was set at p < 0.05 for all statistical analyses.
Figures were made with Adobe Photoshop CS6 and Illustrator CS6, and brightness/contrast, orientations, and background corrections were applied to better illustrate the staining patterns.

Results
Deleting NMNAT2 in post-mitotic cortical glutamatergic neurons results in deformed brains and agedependent loss of long-range cortical axons Germline deletion of NMNAT2 in mice results in premature death at birth with severe axonal outgrowth de cits and subsequent degeneration in peripheral and optic nerves [37,39]. Using explant cultures prepared from NMNAT2 KO embryonic cortices, Gilley et al. showed NMNAT2 KO cortical axons were shorter than control axons and exhibited degenerative phenotypes [37]. The premature death of germline NMNAT2 KO prevented in vivo examinations of the role of NMNAT2 in cortical neurons. Using mRNA in situ hybridization, we found that nmnat2 mRNA is enriched in the embryonic cortical plate, where postmitotic glutamatergic neurons are located ( Fig. 1-S1A). Axons of cortical glutamatergic neurons often travel long distances to the contralateral hemisphere or subcortical regions via extensive and complex arbors [6]. These observations led us to hypothesize that NMNAT2 is required for axonal outgrowth and the health of cortical glutamatergic axons.
To test this hypothesis, we generated NMNAT2 conditional knockout mice (cKO) by crossing NMNAT2 conditional allele mice (NMNAT2 f/f ) [37] with a Nex-Cre transgenic mouse line, which expresses Cre recombinase in post-mitotic glutamatergic neurons mainly in the cortex and hippocampus from embryonic day 11.5 (E11.5) [38] (Fig. 1-S1B). Despite the restricted deletion of NMNAT2, only about 50% of cKO mice survived after birth, with no apparent lethality at E18.5 ( Fig. 1-S1C-D). The surviving cKO mice weighed signi cantly less than their littermate controls from early postnatal ages through adulthood (Fig. 1A). NMNAT2 cKO mice exhibited evident hindlimb clasping (Fig. 1B), ataxia and forelimb circling phenotypes (10 out of 10 mice examined), re ecting motor behavioral de cits similar to those observed in many neurodegenerative mouse models [49][50][51]. cKO brains were visibly smaller than controls from postnatal ages (Fig. 1C). To evaluate gross brain morphology, we examined brain sections along rostral to caudal coronal planes with bright eld microscopy. Compared to the brains of littermate controls, the brains of cKO mice displayed enlarged ventricles, smaller hippocampi, aberrant anterior commissures, a thinner corpus callosum, and a thinner cortex (Fig. 1D). We measured the thickness of the primary somatosensory cortex and found that it was signi cantly reduced in cKO compared to their littermate controls at postnatal day 16/21 (P16/21) and P90 (Fig. 1E). The degenerative brain phenotypes observed in NMNAT2 cKO brains highlight the importance of NMNAT2 in cortical neuronal health.
To distinguish whether brain dystrophy in NMNAT2 cKO is caused by axonal outgrowth de cits versus axonal maintenance failures, we examined the corpus callosum, where the long-range callosal axons of cortical glutamatergic neurons cross the midline on the way to their contralateral targets [8,9]. To facilitate visualization of the corpus callosum, we immunostained the medium-size neuro lament (NF-M), a cytoskeleton protein enriched in axons, in P4/5, P16/21, and P90 cKO and control brains. There was no difference in corpus callosum thickness between control and cKO brains at P4/5 ( Fig. 1F-G). However, a drastic reduction of corpus callosum thickness in cKO mice occurred by P16/21 and persisted until at least P90, the eldest age examined (Fig. 1F-G). Most callosal axons have already nished midline crossing at P4/5 [52,53]. Thus, the normal corpus callosum thickness in cKO mice at P4/5 suggests that NMNAT2 deletion in glutamatergic neurons does not impair initial axonal outgrowth. Instead, the agedependent reduction of corpus callosum thickness and degeneration-like brain morphology suggest that NMNAT2 is required to maintain the health of long-range cortical axons.

NMNAT2 loss leads to APP accumulation in axons prior to degeneration in vivo and in vitro
Based on the axon degeneration phenotype in NMNAT2 cKO brains, we hypothesized that NMNAT2 loss disrupts axonal physiology, ultimately resulting in axonal degeneration. Axonal transport plays a critical role in neuronal function and survival [28] and its de cits are thought to be a primary cause of axonopathy [54]. Amyloid Precursor Protein (APP) is rapidly and bidirectionally transported in axons [55][56][57][58][59] and is required for synaptogenesis and synaptic function, plasticity, etc. [60-62]. Axonal APP accumulation has been found in AD mouse models [63, 64] and traumatic brain injury [65, 66], and serves as a marker for axonal transport breakdown [67].
To examine whether NMNAT2 deletion results in axonal transport de cits, we immunostained APP in brains prepared from P5 and P21 cKO mice and littermate controls. We found signi cant APP accumulation in the corpus callosum at P5 in cKO but not in control ( Fig. 2A, E). This P5 timepoint is before the corpus callosum thickness is affected by NMNAT2 loss. In addition, we observed APP accumulation in cKO mice in brain regions where glutamatergic axons transit, including the hippocampal mbria and striatum. In contrast, little APP signal was detected in the corresponding regions of control brains (Fig. 2B, C, E). At P21, where signi cant axon degeneration occurs, we found a striking buildup of the APP signal in areas enriched with long-range axons. As APP accumulations are encapsulated by myelin basic protein (MBP), a marker for myelinated axons, in the corpus callosum (Fig. 2D), it suggests that the APP aggregates are present in axons.
APP accumulation increased drastically in KO axons from DIV8 to 14 ( Fig. 2-S1A, B), while the signal densities for TUJ1 and MAP2 levels were similar between KO and control neurons ( Fig. 2-S2A, B).
Additionally, we detected fragmented and aggregated TUJ1 signal as a sign of axon degeneration at DIV14 (Fig. 2-S1 and Fig. 8-S2C). As our cortical neuronal cultures are prepared from E15.5/E16.5 embryonic cortex, by DIV8, they likely correspond to the rst postnatal week of age in vivo. Additionally, we found signi cant axonal APP accumulation in NMNAT2-deleted cortical neurons using an alternative Cre-loxP approach (Fig. 2-S3). This in vitro recapitulation of APP accumulation and axon degenerationlike phenotype seen in vivo justi es using cultured NMNAT2 KO neurons as a cellular model to elucidate the molecular mechanisms mediating NMNAT2's function in axonal health.

NMNAT2 is required for the transport of fast-moving vesicular cargos in distal axons
As APP accumulation is observed prior to axonal degeneration in NMNAT2 KO axons, we hypothesized that NMANT2 is required for axonal transport. Previous work showed that NMNAT2 is localized to Golgiderived vesicles and undergoes fast, bi-directional axonal transport in PNS axons [25,26]. We observed a similar transport of NMNAT2 along axons and its colocalization and comigration with fast-moving cargos in cultured cortical neurons (Fig. 3-S1).
To determine whether NMNAT2 is required for axonal transport, we used time-lapse imaging to quantify axonal transport of EGFP-tagged vesicular cargos, APP and SNAP25, and DsRed-tagged mitochondria in control and NMNAT2 KO neurons at DIV6 and 8 ( Fig. 3 and Fig. 3-S2). APP is a component of Rab5containing vesicles [68], while SNAP25 is a component of Piccolo-bassoon transport vesicles [69]. Rab5 and SNAP25 undergo fast, bidirectional axonal transport [70,71] mediated by kinesin-1 and dynein motor proteins [55,72]. Distinct from vesicular cargos, axonal mitochondria move slowly and intermittently in both directions [73,74], responding to the interplay between adaptor and motor proteins [75,76]. To transfect only a few neurons with APP-EGFP, SNAP25-EGFP, or mito-DsRed expressing construct, lipofectamine transfection was conducted < 20 h before live imaging to reduce toxicity of overexpression.
Such sparse labeling allowed us to identify distal (> 400 µm away from the soma) or proximal (within 200 µm of the soma) axon segments (Fig. 3-S3). Axonal transport of APP and SNAP25 was measured in distal and proximal segments (Fig. 3 and Fig. 3-S2). At DIV8, signi cant de cits in APP and SNAP25transport were detected in KO distal segments (Fig. 3) but not proximal segments (Fig. 3-S2). Furthermore, we observed a substantial increase in the percentage of vesicles in the stationary and dynamic pause phases (Fig. 3C, F), a concomitant decrease in the percentage of vesicles engaging in anterograde movement, and reduced velocities in anterograde and retrograde directions (Fig. 3D, G).
However, at DIV4 (data not shown) and 6 ( Fig. 3C, D, F, G), APP and SNAP25 transport were unaffected in KO axons. In contrast to vesicular transport, mitochondrial distribution, morphology, and motility were unaffected in KO neurons at DIV8 (Fig. 3-S4), despite the heterogeneous mitochondrial velocities reported previously [77]. These results demonstrate that NMNAT2 is required for fast transport of vesicular cargos but not mitochondria, and that NMNAT2 loss affects distal axon transport before axon degeneration.
NMNAT2 maintains global NAD, NADH levels, and local NAD/NADH redox potential in distal axons NMNAT2 catalyzes NAD synthesis in the salvage pathway, the major pathway in CNS neurons for NAD biosynthesis [15]. The ratio of oxidized (NAD + ) to reduced (NADH) forms of NAD establishes the NAD redox potential (NAD + /NADH) and is crucial to driving glucose metabolism [78,79]. To determine if NMNAT2 is required for maintaining the NAD redox potential, we measured the abundance of NAD + and NADH at DIV8, when axonal transport de cit rst becomes evident in KO axons (Fig. 2-S1 C,D). Both NAD + and NADH levels were reduced to ~ 50% of their control value in KO neurons, suggesting that NMNAT2 is a major source of NAD in cortical neurons. However, since NAD + and NADH levels were reduced to the same extent upon NMNAT2 loss, the NAD redox potential measured from whole neurons remained unchanged (Fig. 4A).
One reason that NMNAT2 loss affects axonal transport in distal but not proximal axons may be that its homolog NMNAT1, which is expressed in the neuronal nucleus, is su cient to maintain NAD redox potential in the soma and proximal axons [14,80]. Thus, we hypothesized that NMNAT2 is only critical for maintaining NAD redox potential in distal axons, where it provides the ATP required for fast axonal transport. So, to measure NAD redox potential in sub-axonal regions, we used a genetically encoded sensor, SoNar [44], that detects cytosolic NAD redox potential with a reliable signal-to-noise ratio and high spatial resolution. Our imaging studies found that NAD redox potential was signi cantly reduced in distal axons but not in the soma or proximal axons of DIV8 KO neurons compared to controls (Fig. 4B-C). These data con rm the requirement of NMNAT2 in distal axons to maintain NAD redox potential. In addition, the data suggest that NMNAT2 contributes to ~ 50% of the overall NAD + and NADH levels in cortical neurons.

NMNAT2 is critical for glycolysis on synaptic vesicles
Fast axonal transport of vesicular cargo is fueled by on-board glycolysis through vesicle-associated glycolytic enzymes that generate ATP to support motor protein movement [32,81]. Considering the transport de cits and the reduced NAD + /NADH redox potential in the distal axons of NMNAT2 KO neurons, we hypothesized that NMNAT2 supports fast axonal transport by facilitating local glycolysis. To measure ATP near fast-moving vesicular cargos, we transfected the genetically encoded presynaptic ATP sensor (Syn-ATP) in control and NMNAT2 KO neurons and measured ATP levels in distal axons at DIV 8 with live imaging (Fig. 5). This sensor is targeted to synaptophysin, one of the well-known fast vesicular cargos. Syn-ATP comprises an optimized luciferase to detect ATP by luminescence and an mCherry uorescent protein as an internal control for sensor expression level [33] (Fig. 5B). The luminescence to uorescence ratio (L/F) measured from Syn-ATP is proportional to ATP levels near synaptic vesicles (sv-ATP) [33]. As the Syn-ATP sensor is pH sensitive [33], all the measurements were pH-corrected ( Fig. 5-S1A-C).
NMNAT2 KO neurons exhibited a modest but signi cant decrease in sv-ATP compared to control neurons ( Fig. 5C,D). sv-ATP can come from local glycolysis and/or mitochondrial ATP synthesis [82]. To check if it was contributed by glycolysis, we applied oligomycin, an F 1 -F 0 ATP synthase inhibitor that blocks mitochondrial ATP production (Fig. 5A,B). Previous ndings show that glycolytic ATP production is the primary ATP source for fast-transporting cargos [32,81]. However, we found no signi cant reduction in sv-ATP levels (p = 0.3621) in control distal axons upon oligomycin treatment (Fig. 5C,D). However, in KO distal axons, oligomycin treatment signi cantly and strongly reduced sv-ATP levels (Fig. 5C,D), suggesting that in the absence of NMNAT2, mitochondria provide ATP in distal axons. These data strongly indicate that NMNAT2 is required to drive glycolysis on synaptic vesicles.
To test if the reduced sv-ATP in NMNAT2 KO axons is caused by NAD + de ciency, neuronal cultures were supplemented with 1 mM NAD + from DIV5 to DIV8 (Fig. 5A). Isotope labeling studies show that primary neuronal cultures can take up exogenous NAD + [83, 84]. Here we found that NAD + supplementation increased intracellular NAD + levels transiently in neuronal cultures (Fig. 5-S2) and thus we refreshed NAD + daily. NAD + supplementation restored sv-ATP levels in KO distal axons to control levels in both basal and oligomycin treated conditions (Fig. 5C-D). These data suggest that NMNAT2 synthesizes the NAD + required to drive glycolysis on synaptic vesicles. NAD supplementation restores APP transport via glycolysis in the absence of NMNAT2 Next, we determined whether NAD + supplementation can restore fast axonal transport in NMNAT2 KO distal axons. Compared to vehicle treatment, NAD + supplementation signi cantly decreased the percentage of stationary/dynamic pause events, increased the percentage of anterograde and retrograde events, and restored anterograde and retrograde velocities of APP transport (Fig. 6A-C). On the other hand, NAD + supplementation of control neurons had minimal impact on APP axonal transport (Fig. 6A-C) and NAD + abundance (Fig. 5-S2).
We then tested whether glycolysis is required for NAD + rescue of APP transport. We acutely suppressed glycolysis by transferring the neurons to a glucose-free medium containing the hexokinase inhibitor, 2deoxyglucose (2DG), while adding methyl-pyruvate (Methyl-pyr) as an alternative substrate to support TCA cycle and OXPHO in mitochondria (Fig. 6D,E). In parallel, we tested if mitochondrial ATP production is required for NAD + rescue of axonal transport in NMNAT2 KO neurons by blocking OXPHO in the presence of glucose (Fig. 6D,E). Glycolysis inhibition signi cantly impaired APP transport in control axons, as revealed by a signi cant increase in stationary/dynamic pause events and decrease in transport velocity (Fig. 6F,G). In fact, the substantially reduced mobile APP-tagged vesicles in glycolysisinhibited cultures made it challenging to image su cient vesicles for velocity measurements. In contrast, OXPHO inhibition had a relatively mild impact on APP transport in control axons, including a moderate reduction in anterograde velocity and a minor but signi cant increase in the percentage of retrograde events. In KO axons, glycolysis inhibition abolished the rescue of APP transport provided by exogenous NAD + and signi cantly reduced the number of transport events and movement velocities (Fig. 6F,G).
OXPHO inhibition also perturbed NAD + -mediated rescue of KO neurons, although not as robustly as did glycolysis inhibition, as shown by the increased percentage of stationary/dynamic pause events and decreased percentage of anterograde events (Fig. 6F). However, the transport velocities in both directions were unaffected (Fig. 6G). Additionally, we assessed APP accumulation by immunostaining and found that NAD + supplementation from DIV8 to 14 signi cantly reduces APP accumulation in KO neurons.
Furthermore, 48 hours of glycolysis inhibition together with methyl-pyruvate supplementation abolished this rescue (Fig. 6-S1A-C). These data demonstrate that NAD + -mediated rescue of APP axonal transport in NMNAT2 KO neurons depends mainly on glycolysis with a modest contribution from mitochondrial OXPHO. SARM1 depletion sustains APP transport and prevents axon degeneration in the absence of NMNAT2 Sterile alpha and TIR motif-containing protein 1 (SARM1) is a recently discovered NAD(P) glycolhydrolase highly expressed in neurons [85,86]. SARM1 senses the rise in nicotinamide mononucleotide (NMN)-to-NAD + ratio caused by loss of NMNAT2 and responds by activating its NAD + hydrolase domain [87,88]. SARM1 deletion in NMNAT2 KO mice prevents perinatal lethality, preserves healthy axons and synapses in the peripheral nervous system, and maintains motor function throughout adulthood [89, 90].
We, therefore, examined if SARM1 reduction could reverse axon phenotypes in NMNAT2 cKO brains. To this end, we crossed NMNAT2 cKO mice to SARM1 KO (S null /S null ) mice to generate NMNAT2 cKO missing one copy of SARM1 (NMNAT2 cKO; S null /+), which we backcrossed to SARM1 KO mice to obtain NMNAT2 cKO missing both copies of SARM1 (NMNAT2 cKO; S null /S null ). NMNAT2 cKO; S null /+ and NMNAT2 cKO; S null /S null mice survived similarly to their littermate controls, in contrast to NMNAT2 cKO mice (data not shown). Normal brain morphology was observed in NMNAT2 cKO; S null /S null mice (Fig. 7A). In contrast to NMNAT2 cKO; S null /S null mice and control mice, NMNAT2 cKO; S null /+ mice still exhibited reduced body weights (Fig. 7-S1B). Gross examination of their brains found aberrant anatomy similar to NMNAT2 cKO brains (Fig. 7A), including atrophied hippocampi, enlarged ventricles, and thinned primary somatosensory cortex (Fig. 7B) and corpus callosum (Fig. 7C,D). APP accumulation was found in the corpus callosum, mbria, and striatum in NMNAT2 cKO; S null /+ but not in NMNAT2 cKO; S null /S null brains (Fig. 7E,F). APP accumulation was detectable at P5 in NMNAT2 cKO; S null /+ but not NMNAT2 cKO; S null /S null brains, suggesting APP transport was already normalized by the complete absence of SARM1 at P5 (Fig. 7-S2). Taken together, these ndings show that complete SARM1 loss prevents the impact of NMNAT2 loss in cortical axons.
Next, we examined whether SARM1 loss protects NMNAT2 KO neurons from axonal degeneration by preventing axonal transport de cit. We used anti-sense oligonucleotides targeting sarm1 mRNA (SARM1-ASO) to knockdown SARM1 expression at desired time points and a scrambled anti-sense oligonucleotide as the non-targeting control (ctrl-ASO). We evaluated the SARM1 knockdown e cacy of two SARM1-ASOs and one ctrl-ASO using qPCR and western blotting. ASO33, one of the two SARM1-ASOs, signi cantly decreased SARM1 protein abundance 4-7 days following ASO treatment (Fig. 8-S1).
This ASO was therefore used for the rest of the experiments and referred to as SARM1-ASO. To test the impact of SARM1 knockdown on axonal transport, we added ASOs from DIV1 and DIV5 on and examined the impact at DIV8. SARM1-ASO application starting at DIV1 signi cantly reduced SARM1 abundance bỹ 70% and prevented APP transport de cits in NMNAT2 KO axons at DIV8 (Fig. 8A1-4). In contrast, SARM1-ASO treatment starting at DIV5 only reduced SARM1 abundance by ~ 50% and failed to rescue axonal transport (Fig. 8B1-4). Surprisingly, by DIV12, APP transport was completely rescued in the distal axons of these neurons (Fig. 8C1-4), despite the impairment at DIV8. Consistent with axonal transport analysis, APP accumulation at DIV14 was rescued by SARM1-ASO treatment starting from DIV1 or DIV5 in NMNAT2 KO neurons (Fig. 8-S2). Furthermore, the axon degeneration phenotype revealed by TUJ1 aggregates at DIV14 in ctrl-ASO-treated NMNAT2 KO axons was also reduced by SARM1-ASO treatment.
Thus, blocking NAD + degradation by SARM1 depletion protects axons during NMNAT2 loss in vivo and in vitro.

Discussion
NMNAT2 has been identi ed as an AD target and an axonal maintenance factor. Here we provide the rst in vivo evidence that NMNAT2 is critical for the health of long-range cortical glutamatergic axons in mice. We show that NMNAT2 loss impairs glycolysis, disrupts fast axonal transport, and results in APP accumulation. NAD + supplementation or reducing the levels of SARM1, an NAD + hydrolase, effectively restores fast axonal transport and prevents the neurodegeneration commonly observed in NMNAT2 cKO axons both in vitro and in vivo. In summary, our present study suggests that NMNAT2 protects cortical neurons from axonal degeneration by ensuring that the energetic demands of distal axons are met. Our data also suggest glucose hypometabolism in long-range axons is likely a major cause of axonopathy during NDAs. Therefore, therapies in preventing NMNAT2 or NAD loss during aging or NDAs should maintain axonal energetics and axonal health.
NMNAT2 contributes to fast axonal transport by maintaining the local NAD redox potential for e cient vesicular glycolysis In this study, we show that NMNAT2 is required for maintaining vesicular glycolytic activity (Fig. 5) and the fast axonal transport of APP and SNAP25-containing vesicular cargos (Fig. 3) in distal axons of cortical neurons. This role in vesicular cargo transport is consistent with earlier observations. For instance, NMNAT2 is enriched in the synaptic vesicle and membrane but not in mitochondrial fractions of cortical neurons [91]. In addition, the palmitoylation of NMNAT2 enables it to associate with the membrane of Golgi-derived, fast-moving vesicular cargos [25]. Our time-lapse imaging con rms that NMNAT2's dynamic antero-and retrograde movement and transport velocities in axons are similar to fast-transported cargos like APP-and SNAP25-containing vesicles (Fig. 3-S1).
Glycolysis is the primary ATP source for fast axonal transport and is generated by a vesicular glycolytic complex [32,81]. It has been estimated that one kinesin requires ~ 187.5 ATP molecules/s to move at ~ 1.5 µm/s [36], the average anterograde velocity of APP. Thus, ~ 188 NAD + molecules of NAD + are needed to drive ~ 94 glycolytic reactions per second near individual vesicular cargos to generate the ATP consumed by one kinesin. Data from our SoNar NAD + /NADH radiometric imaging studies show that NMNAT2 is essential in maintaining the NAD redox potential in distal axons. Based on in vitro biochemical studies, one NMNAT2 synthesizes ~ 0.6 NAD + per second [91]. Knowing that lactate dehydrogenases (LDH) can catalyze NADH recycling back to NAD + , we hypothesize that NMNAT2 coordinates with LDH to maintain NAD redox potential for glycolytic ATP production. Furthermore, our data suggest that NMNAT2's NAD + synthesizing activity and proximity to the vesicular glycolytic complex ensure su cient ATP for motor protein operation in fast axonal transport.
Intriguingly, before DIV8, NMNAT2 seems to be dispensable for axonal extension and axonal transport, which also requires glycolytic ATP supply [92]. We speculate that other mechanisms, such as the newly discovered Wnk kinases [93], exist in the early developmental stage to ensure NAD homeostatic balance in the absence of NMNAT2, allowing a su cient level of glycolytic ux to fuel cargo transport, cytoskeleton dynamics, and axon outgrowth. However, by DIV8 in vitro and P16/21 in vivo, NMNAT2 becomes the mandatory source of NAD + , but only in distal axonal segments. In proximal axons, NMNAT2 is dispensable for NAD redox potential maintenance and fast axonal transport. Given that the NAD + pool between the nucleus and soma is interchangeable [45], proximal axons could potentially receive su cient NAD + generated by nuclear NMNAT1 due to their proximity to soma and thus maintain an adequate NAD redox potential to drive proximal axonal transport despite the absence of NMNAT2. Alternatively, OXPHO from somatic mitochondria could fuel axonal transport in proximal axons of NMNAT2 KO neurons.
Electron microscopy studies of cortical layers 2/3 of the primary visual cortex show that the proximal axons within 100 µm from somata of pyramidal neurons tend to have higher mitochondrial coverage and volume than distal axons [94]. Mitochondria tend to be smaller, and less mobile in distal axons [94][95][96][97].
As mitochondria are smaller and less mobile in distal axons [94][95][96][97], we hypothesize that higher coverage of mitochondria in proximal axons compensates for the glycolysis de cits in NMNAT2 KO neurons to fuel fast axonal transport.
Previous studies indicate that BDNF axonal transport is mainly fueled by glycolysis, and unaffected by ablating mitochondrial OXPHO [32]. Our data is largely consistent with this nding on glycolysisdependent fast axonal transport. However, upon OXPHO inhibition, we observed a mild but signi cant increase in the numbers of APP cargos undergoing retrograde transport, with a concomitant decrease in anterograde velocity. Using Drosophila neurodegenerative models, it has been shown that mitochondrial dysfunction stimulates a retrograde signaling response [98]. APP is partially localized on the endolysosomal vesicles involved in retrograde transport [99,100]. Thus, the increase in retrograde APP cargo upon mitochondrial inhibition can re ect enhanced retrograde signaling.
Accumulation of APP, a pathological hallmark of defective axonal transport [67,101], was found in regions enriched with long-range axons of NMNAT2 cKO brains already at P5, preceding the obvious axonal loss. No APP accumulation was observed in the cortex, where proximal or short axonal arbors are enriched. Using cultured neurons, we demonstrated that the fast axonal transport of APP and SNAP25 is severely impaired in distal axonal segments of NMNAT2 KO neurons. These in vitro observations of speci c axonal transport de cits in distal axons of NMNAT2 KO cortical neurons offer a mechanistic explanation for the APP accumulation in long-range axons observed in cKO brains. Together, our study provides strong evidence that NMNAT2 is required for fast axonal transport in distal axons of cortical glutamatergic neurons.
Cumulative evidence suggests that neuronal subtypes with long and complex axonal arbors are particularly vulnerable to external insults [102,103]. As NMNAT2 cKO brains reach the age of P16-P21, not only does APP accumulation drastically increase, but severe axon degeneration occurs, along with a reduction in cortical thickness, hippocampal atrophy, ventricle enlargement, and motor behavioral de cits, recapitulating major hallmarks of neurodegeneration. These ndings support the hypothesis that defective axonal transport drives axonopathy in the pre-symptomatic phase of neurodegeneration [54,104], which may serve as the focal basis for the gradual development and spread of secondary neuronal damage [105,106]. However, we do not exclude a possible role of NMNAT2 in maintaining neuronal health in the dendritic and somatic compartments [107,108], whose dysfunction could also contribute to neurodegenerative phenotypes in NMNAT2 cKO brains.
Mitochondrial OXPHO as a mediator of axonal degeneration in NMNAT2 KO cortical neurons Our sv-ATP imaging studies show signi cantly reduced ATP levels in NMNAT2 KO distal axons when mitochondrial function is blocked (Fig. 5). By contrast, in control cortical axons, mitochondrial inhibition has a negligible impact on sv-ATP levels. These observations suggest that NMNAT2 loss results in a shift towards mitochondrial OXPHO to maintain sv-ATP levels, however, not enough to fuel fast axonal transport. The literature suggests two mechanisms that could mediate this shift in CNS neurons: (1) Ca 2+ signaling-dependent regulation of mitochondrial calcium uniporters and the malate-aspartate shuttle [109][110][111]; (2) The O-GlcNAcylation-dependent post-translational modi cation of the mitochondrial proteome and mitochondrial motility [112,113]. It remains unclear why this metabolic shift fails to protect NMNAT2 KO neurons from axonal degeneration: at DIV14, NMNAT2 KO axons exhibit prominent blebbing, a degenerative-like phenotype (Fig. 8-S2B). It has been suggested that the cumulative oxidative stress and depletion of TCA cycle substrates from hyperactive mitochondrial OXPHO could be detrimental to neuronal survival in PD and AD animal models [114][115][116]. This observation raises the possibility that excessive mitochondrial activity, compensating for defective glycolysis in NMNAT2 KO axons, contributes to axon degeneration.
We suspect SARM1 depletion not only mitigates the harmful impact of excessive OXPHO, but also sustains OXPHO by preventing mitochondrial NAD + pool decline, thus offering excellent rescue in NMNAT2 KO neurons. Endogenous SARM1 widely distributes along neurites as small puncta [47,117].
Upon mitochondrial stress or overexpression, SARM1 localizes onto the mitochondrial outer membrane and interacts with mitophagy machinery [85,86,118,119]. SARM1 is a NAD + consuming enzyme and its NAD + hydrolase activity can be activated by the rise of NMN to NAD + ratio [87], or by JNK-mediated phosphorylation [120]. It has been shown that SARM1 in activated state drastically consumes NAD + and impairs mitochondrial OXPHO [120]. In an axotomy model, the mitochondrial motility is preserved when SARM1 is genetically deleted [121]. In the Charcot-Marie-Tooth neuropathy model with mitochondrial abnormality, SARM1 deletion preserved mitochondrial morphology and motility, and protected axons and synapses from degeneration [122]. Here we nd that SARM1 loss prevents several phenotypes caused by NMNAT2 loss, including the impaired fast axonal transport and axonal morphology. Currently, it is still unknown whether SARM1 removal in NMNAT2 KO axons prevents the oxidative stress caused by mitochondria hyperactivity. Further investigations into the impacts of SARM1 deletion from NMNAT2 KO axons on mitochondrial metabolic capacity, dynamics, NAD + -NADH shuttling between mitochondrial matrix and cytoplasm, and mitochondrial quality control are needed.
Anatomically, these hub areas are densely inter-connected by long-range axonal tracts [137], that often show pathological changes since early-stage disease [138]. Metabolically, these connectome hubs express higher levels of glycolysis genes compared to the non-hub areas [139], and exhibit reduced aerobic glycolysis upon normal aging and AD [140,141], which positively correlates with white matter deterioration [142]. Notably, glucose hypometabolism is observed directly in white matter tracts of AD brains [4].
We show that NMNAT2 is critical for vesicular glycolysis to fuel axonal transport in distal axons. Previously, we have demonstrated that NMNAT2 levels are signi cantly reduced in the prefrontal cortex of AD brains [21], one of the DMN hubs. Our biochemical assays reveal a ~ 50% reduction of NAD + and NADH levels in NMNAT2 KO neurons (Fig. 4A), indicating that NMNAT2 is a major NAD + and NADH provider in cortical neurons. Synthesizing these observations, we hypothesize that reduced NMNAT2 in AD brains contributes, at least in part, to the reduction in glucose metabolism observed in AD white matter.
NMNAT2 reduction has also been observed in PD and HD brains [21][22][23]. PD patients show hypometabolism in the premotor and parieto-occipital cortex that correlates with motor dysfunction [128], while HD patients show progressive glucose hypometabolism in the frontal lobe, temporal lobe, and striatum, accompanied by white matter volume reduction [125]. Supplementing NAD + and its precursors remarkably ameliorates degenerative phenotypes in transgenic AD and ALS mouse models [143][144][145].
Similarly, increasing NMNAT2 expression broadly provides neuroprotection across mouse models of tauopathy [21,146], familiar AD [147], and glaucoma [148]. Coincidently, upregulating glycolysis exerts neuroprotective effect in PD synucleinopathy and ALS models [149,150]. Our study highlights a novel role of NMNAT2 in supporting glycolysis in long-range projecting axons of cortical glutamatergic neurons. Extrapolating from our ndings, NMNAT2 could serve as a putative therapeutic target to boost neuronal glycolysis in order to antagonize the structural connectome breakdown occurring in many neurodegenerative disorders.     Loss of NMNAT2 reduces NAD + levels and impairs the NAD redox potential in distal axons (A) Levels of NAD + and NADH and calculated NAD + /NADH ratios in DIV8 ctrl and KO cortical neurons as measured by the NAD + /NADH-Glo assay and normalized to protein amounts. Readings from 28 ctrl and 28 KO culture wells from 4 independent culture experiments; unpaired t-test was used for NADH while  Deleting SARM1 prevents neurodegenerative phenotypes in NMNAT2 cKO brains.

Abbreviations
(A) Bright eld images of coronal brain sections (rostral to caudal from left to right) from ctrl, NMNAT2 cKO; S null /+, and NMNAT2 cKO; S null /S null mice. Enlarged ventricle, greatly reduced corpus callosum, and internal capsule were evident in NMNAT2 cKO; S null /+ but not in NMNAT2 cKO; S null /S null brains. (B)