NMNAT1-overexpressing cells inhibit SARM1 activation in neighboring cells through gap junction intercellular communication.
We first verified the metabolic regulation of SARM1 is operational in cells. SARM1 activity was monitored by measuring its enzymatic product, cADPR, following activation by CZ-4821. Overexpressing the nuclear NMNAT133 or Golgi-localized NMNAT233, elevated NAD levels in HEK-293T cells and, indeed, inhibited SARM1 activation. The mitochondrial NMNAT333 was less effective (Fig. 1a and Supplementary Fig. S1a-b). To determine whether the NAD synthesis activity of the NMNAT1 was responsible for the inhibition, we mutated several residues within or surrounding the catalytic pocket of NMNAT1 (Supplementary Fig. S1c) to glycine. These mutants exhibited different activities in inhibiting the CZ-48-induced cADPR production in HEK-293T cells (Supplementary Fig. S1d, red dots), which was negatively correlated with their NAD synthetase activities (Supplementary Fig. S1d, blue dots), indicating that NAD synthesis activity was indeed responsible. It is in consistent with the report that NAD binds to the ARM domain and mediates self-inhibition of SARM122. Conversely, knocking out NMNAT1 elevated cellular NMN and activated SARM1, which was reversed by re-expression of NMNATs (Supplementary Fig. S1e), verifying that the metabolic regulation of SARM1 by the cellular levels of NMN and NAD21, 24, 34 is indeed operational in cells. The less effectiveness of NMNAT3 may be due to the low level of mitochondrial transporter of NAD, SLC25A51 in HEK-293T35, hampering the transfer of the mitochondrial NAD to the cytosol.
In the above experiments, even with only approximately 50% of cells showing positive NMNATs staining (Supplementary Fig. S1a), complete inhibition of cADPR production in the entire culture was observed (Fig. 1a), suggesting possible intercellular transfer of inhibition. To quantitatively document this effect, we mixed different ratios of wildtype HEK-293T cells expressing low levels of NMNATs with NMNAT1-overexpressing stable cell lines. Treatment with CZ-48 induced cADPR production in wildtype HEK-293T cells but not in NMNAT1-overexpressing cells (Fig. 1b, left 4 columns). Notably, the 1:1 mixture displayed no cADPR production, and 10:1 mixture exhibited approximately one-third of the cADPR amount compared to the HEK-293T group (Fig. 1b, right 2 columns). These results demonstrate that NMNAT1 not only inhibits CZ-48-induced SARM1 activation and subsequent cADPR production in its own cells but also in neighboring cells in the culture.
We further investigated this phenomenon using a different system, by mixing NMNAT1-KO HEK-293T cells with wildtype HEK-293T cells. The NMNAT1-KO cells displayed remarkably high levels of cADPR, shown also in Supplementary Fig. S1e, much higher than the wildtype HEK-293T cells. Upon mixing the two cell types in a 1:1 ratio, only minimal cADPR production was observed (Fig. 1c). This finding suggests that the endogenous NMNAT1 in wildtype HEK-293T cells can inhibit NMN-induced SARM1 activation in NMNAT1-KO cells.
Since cells communicate through various mechanisms, we sought to determine whether the above phenomenon requires cell-cell contact. We conducted additional experiments to address this question. Firstly, we seeded the two cell lines separately in two chambers of a Boyden Chamber and measured the cellular cADPR levels in the lower chamber. In this configuration, the wildtype HEK-293T cells did not inhibit cADPR production in NMNAT1-KO cells (Fig. 1d). Secondly, we cultured NMNAT1-KO cells with the conditioned medium of the two cell lines. Interestingly, the conditional medium from wildtype HEK-293T cells did not inhibit cADPR production in NMNAT1-KO cells (Fig. 1e). These results strongly suggest that direct cell-cell contact is required for the intercellular regulation of SARM1 activation observed in our experiments.
To further explore the role of cell-cell contact, we co-seeded an equal number of the 1:5 mixed HEK-293T cells expressing EGFP and NMNAT1-KO cells in wells with different surface areas, specifically a 24-well plate and a 6-cm dish, resulting in approximately a 10-fold difference in cell density (Fig. 1f). In the 24-well plate, where all cells formed contacts with other cells, the healthy HEK-293T cells (EGFP positive) efficiently alleviated NMNAT1-KO-induced NMN accumulation and NAD depletion, subsequently inhibited cADPR production in the mixed population (Fig. 1g, blue dots). These findings are consistent with our previous observations (Fig. 1a-e). In contrast, in the 6-cm dish, where around half of the population grew as single cells without cell-cell contacts, the mixture showed similar trends in nucleotide levels, but to a significantly lesser extent (Fig. 1g, red dots). This suggests that the inhibitory effect quantitatively depends on the presence of cell contacts.
To investigate the potential involvement of gap junctions in mediating the observed inhibitory effect, we performed RNA sequencing of HEK-293T cells, which revealed relatively high transcriptional levels of three connexins, including connexin 43 (Cx43), connexin 45 (Cx45) and connexin 46 (Cx46). We subsequently employed CRISPR/cas9 technology to knockout these genes in HEK-293T cells. The knockout cells were validated through Western blot analysis using specific antibodies against connexins (Fig. 1h) and then mixed with NMNAT1-KO cells. Interestingly, Cx43-KO cells exhibited a significantly reduced ability to replenish the NAD store in NMNAT1-KO cells compared to wildtype HEK-293T cells, while knockout of the other two connexins did not affect the inhibitory effect (Fig. 1i). These data suggest that the intercellular regulation of SARM1 activation among HEK-293T cells occurs through gap junctions, predominantly by Cx43.
Visualization of intercellular communication of nucleotides via gap junctions.
Previously, we have developed a series of fluorescent probes that are not only highly sensitive to measure the activity of SARM1 in vitro but also capable of imaging the activity in live cells induced by CZ-48, vincristine, or axotomy36, 37. Here, we employed PC11 (Probe 1a in literature37) to visualize intercellular nucleotide communication as well. PC11 undergoes a catalytic base exchange with the nicotinamide (Nam) group in the NAD molecule. The reaction is selectively catalyzed by NMN- or CZ-48-activated SARM1, resulting in the formation of a fluorescent analogue of NAD called PAD11 (Fig. 2a). It is a highly charged molecule and is thus cell impermeant and can only be transferred via GJs to other cells. By monitoring the fluorescence of PAD11, we can visualize its intracellular transfer, and in inference, that of nucleotides with similar size and charges, such as NAD, NMN or Nam.
HEK-293 cells overexpressing SARM1, treated with 100 µM CZ-48 and 12.5 µM PC11 for 8 h, exhibited a bright orange fluorescence due to the accumulation of PAD11 (Fig. 2b, second panel). In contrast, wildtype HEK-293 cells, expressing low levels of endogenous SARM121 and exogenous mCherry as a marker, did not display observable PAD11 signals (Fig. 2b, first panel). Interestingly, when these two stable cell lines were mixed in a 1:1 ratio, all the cells within the same cluster, comprising both mCherry-positive (wildtype) and mCherry-negative (SARM1-OE) HEK-293 cells, exhibited similar intensities of PAD11 signals (Fig. 2b, third panel). We observed three distinct types of cell clusters in the same field of the mixture (Supplementary Fig. S2a), where the mCherry-positive cluster did not show PAD11 signals (red square), the mCherry-negative cluster exhibited PAD11 fluorescence (orange square), and the pink squared cluster consisted of mCherry-positive cells acquiring PAD11 through contact with SARM1-OE cells. These findings confirmed that nucleotides can indeed travel among cells.
Quantification of the same experiments using flow cytometric analyses is shown in Fig. 2c. Activation by CZ-48 for 4 hrs of SARM-OE greatly increased PAD11 fluorescence (Fig. 2c, left panel). Very low PAD11 fluorescence was seen in wildtype (Fig. 2c, middle panel, mCherry positive) due to low expression of endogenous SARM1. When the two cell types were mixed one to one, PAD11 was seen in all cells (Fig. 2c, right panel), even though the PAD11 fluorescence was lower than in the culture of SARM1-OE alone (Fig. 2c-d). This suggested that the PAD11 molecules produced by SARM1-OE cells were diluted by wildtype HEK-293 cells and confirmed the transfer of nucleotides between cells.
To rule out the possibility of SARM1 transfer with mitochondrion through tunneling nanotubes38, we stained the mixed cells with a SARM1 antibody. The results showed exclusive presence of SARM1 signals in mCherry-negative cells (Supplementary Fig. S2b).
To further validate the involvement of Cx43 in intercellular communication, we performed co-culture experiments using NMNAT1-KO HEK-293T cells and either Cx43-KO cells or wildtype cells as control, in the presence of PC11. In the control co-culture, PAD11 fluorescence levels became similar between the two cell populations after 4 h (Fig. 2e, left panel). However, this equalization process was significantly impaired in the Cx43-KO co-culture (Fig. 2e, middle panel), and the normal communication was restored upon re-expression of Cx43 (Fig. 2e, right panel). The expression of Cx43 in the cells was confirmed by Western blot (Supplementary Fig. S2c). These findings provide additional evidence supporting the role of Cx43 in facilitating nucleotides trafficking among cells.
Interestingly, we observed that PAD11 did not transfer between two different cell types, HEK-293 and SH-SY5Y (Fig. 2f and 2g, left panels). We reasoned that this could be attributed to the differential expression of connexin isoforms. According to the RNAseq data, HEK-293 cells predominantly express Cx43; while SH-SY5Y cells mainly express Cx45. Consistent with this, overexpression of Cx43 in SH-SY5Y cells facilitated intercellular communication of PAD11 between the two different cell types (Fig. 2f and 2g, right panels). The expression of Cx43 in the cells was confirmed by Western blot (Supplementary Fig. S2d and S2e). These findings suggest that the expression of the same connexin isoforms expressing in adjacent cells mediates intercellular nucleotide exchange.
Visualization of inter-neuronal gap junction communication of nucleotides.
SARM1 is a crucial NAD-utilizing enzyme primarily expressed in neurons, playing a key role in axon degeneration8, 19. Results and characterization of its intercellular regulation in the somatic HEK-293 cells provide the needed information to extend the study to neurons. To address this, we established co-cultures of dissociated dorsal root ganglion (DRG) neurons isolated from wildtype or SARM1-knockout mice. Two cell aggregates, representing WT or KO neurons, were positioned in separate corners of the well (Fig. 3a, orange circles in the illustration) and allowed to develop axonal projections until they made contacts (blue arrows in the illustration). Subsequently, the cultures were treated with PC11 in the presence or absence of the activator CZ-48, and after 16 h, the areas prior to contacts zone (red dotted square in the illustration) were imaged using a confocal microscope. Consistent with our previous observation 36, 37, PC11 stained the axons derived from wildtype DRG neurons with CZ-48 treatment, but not those from SARM1-knockout neurons (Fig. 3a, first two rows) or without CZ-48 treatment. However, when SARM1-knockout neurons were co-cultured with wildtype neurons and axonal contacts were established, the SARM1-KO axons also exhibited the orange fluorescence indicative of PAD11 (Fig. 3a, third row). Quantification analysis clearly demonstrated that SARM1-knockout axons in contact with wildtype axons displays a similar amount of PAD11 fluorescence compared to wildtype axons, in contrast to the lack of staining when in contact with knockout axons (Fig. 3b).
To confirm that this communication is dependent on neuronal contacts, we performed time-series imaging. Prior to contacting with wildtype axons, the SARM1-knockout axons exhibited no PAD11 signals (Fig. 3c, upper panel). However, once contacts were formed, they displayed similar brightness of PAD11 signals to the neighboring wildtype axons (Fig. 3c, lower panel).
Connexin 36 (Cx36) is the primary connexin expressed in neurons39. In the co-culture system of wildtype and SARM1-KO neurons, the addition of the Cx36 inhibitor mefloquine (Mef) significantly delayed the communication of PAD11, which was synthesized in wildtype axons induced by CZ-48 (Fig. 3d) or axotomy (Fig. 3e).
To further support the role of Cx36, we constructed an AAV-PHP.eB virus carrying an mScarlet expression cassette as a reporting gene, along with the shRNA targeting Cx36 gene (KD), or a scramble sequence as a control. Neurons infected with Cx36-KD virus exhibited significantly lower PAD11 signals in SARM1-KO cells compared to those infected with Scramble viruses (Fig. 3f-g). Western blot analysis confirmed efficient knockdown of the Cx36 gene in the Cx36-KD group (Fig. 3g, lower panel), confirming that Cx36 mediates GJIC of nucleotides in DRG axons.
Inter-neuronal communication regulates axon degeneration.
The possibility that healthy axons can exert protective effects on adjacent injured ones, through the intercellular transfer of NAD and its metabolites, was investigated. Wildtype DRG neurons, with SARM1-KO DRG neurons as control, were seeded as two separated clusters and cultured until their axons made contacts. The axons on the left side were cut, while those on the right side remained uninjured (Fig. 4a, 2nd row). Additionally, cultures without any cut (Non-cut) and with double cuts served as negative and positive controls, respectively (Fig. 4a, 1st and 4th rows). After 24 hours, the images were captured and degeneration indices were calculated. Comparison of the injured axons from wildtype mice with the uninjured ones revealed fragmentation in the former (Fig. 4a-b, L2/L7 vs L1). However, when injured axons formed contacts with healthy axons, the degeneration index significantly decreased (Fig. 4a-b, L2 vs L7), demonstrating the protective effect of the adjacent healthy neurons on the injured ones. Treatment with the Cx36 blocker, Mef, dramatically impaired this protective effect, leading to similar levels of degeneration in the injured axons to the double-cut group (Fig. 4a-b, L5 vs L2/L7), while uninjured axons remained intact (Fig. 4a-b, R5 vs R2/R4), highlighting the contribution of gap junctions-mediated communication in protection by adjacent healthy axons. The SARM1-KO DRG neurons maintained their integrity in both axotomy and Mef treatment (Fig. 4a-b, L3 and L6).
We thus further validate the role of Cx36 using a knockdown approach, employing the same AAV virus as used in Fig. 3f. DRG neurons from wildtype mice were seeded as two separate islands and infected with the AAV virus. Once the axons made contacts, axotomy experiments were conducted similar to those described in Fig. 4a. Successful virus infection was confirmed by the presence of mScarlet fluorescence in the axons (Supplementary Fig. S3). The fragmentation of the injured axons with Cx36 knocked down was significantly accelerated compared to axons infected with the virus expressing scramble shRNA (Fig. 4c-d, L5 vs L2). The fragmentation was mediated by SARM1, as demonstrated by the intact morphology of SARM1-KO axons (Fig. 4c-d, L6 vs L3). These knockdown experiments faithfully reproduced the results of treatment with the Cx36 blocker, Mef, strongly suggesting that adjacent healthy axons can protect cut axons from degeneration through intercellular communication via Cx36.
The findings highlighted the role of intercellular nucleotide communication in modulating SARM1 activity, thereby enabling healthy neurons to protect adjacent injured neurons. This novel mechanism of SARM1 regulation and neuroprotection is summarized in Fig. 4e.
Cx36 protects neurons from neuroinflammation, SARM1 activation and cell death
It is well documented that GJ are abundantly expressed in neurons, especially in cell bodies and dendritic regions40, 41. Intercellular communication via metabolite transfer as described above most likely occurs in vivo as well, exerting similar protective effect against AxD. To evaluate the protective role of GJIC in mouse brain, we proceeded to knockdown Cx36 and assessed the effects. This was accomplished by using the PHP.eB AAV vector, carrying a Cx36-specific shRNA sequence or a scramble sequence as a comparative control. This vector, a derivative of AAV9, exhibits enhanced CNS tropism, facilitating gene delivery in vivo across the blood-brain barrier42. The virus was administrated via the tail vein infusion into both wildtype and SARM1-KO mice. The reason we chose the viral shRNA technique to knockdown Cx36 in mouse brain instead of direct knockout of the gene was to avoid the well documented compensatory mechanisms likely developed in knockout mice. This was of particular concern since there are many alternate forms of Cx in addition to Cx36.
Three weeks post-infusion, brain tissues from mice were analyzed by immunostaining or Western blot. Infection (the mScarlet fluorescence) was predominantly observed in the brain (Fig. 5a and Supplementary Fig. S4, red) and peripheral nerves (Supplementary Fig. S5a, red), but much less so in other organs (Supplementary Fig. S5b, red), as expected. Immunostaining and Western blots confirmed the presence of Cx36 throughout the brains of the control mice, including in the cortex, cerebellum and brainstem (Fig. 5b), as well as the sciatic nerves (Supplementary Fig. S5a, green), but was markedly diminished in Cx36-KD mice (Fig. 5a and Supplementary Fig. S4, green). Cx36 staining did not yield conspicuous signals in the lung and pancreas (Supplementary Fig. S5b, green).
The most prominent abnormality observed in Cx36-KD mice was neuroinflammation, as indicated by immunostaining of the microglial activation marker, Iba-143, which level was markedly increased as compared with that in the control mice infused with virus containing a scramble shRNA (Fig. 5c, green). The results indicated that the neuroinflammation was specific for Cx36 knockdown and not just due to viral infection. Inflammation was apparently confined to the nervous system, as indicated by the unchanged levels of blood inflammation markers such as IL-6 and IFN-γ (Supplementary Fig. S5c-d). The results show that GJIC via Cx36 plays an important role in protecting the brain from neuroinflammation.
The Cx36-KD-induced neuroinflammation was observed in both wildtype and SARM1-KO mice, as indicated by the statistically similar levels of Iba-1 (Fig. 5d). However, the subsequent inflammation induced cell death was critically dependent on SARM1, which activity was measured by the levels of cellular cADPR, its enzymatic product21, 44. Cx36-KD virus not only induced neuroinflammation but also activated SARM1. In the cortex, cerebellum, and brainstem of the Cx36-KD mice, the cellular levels of cADPR were substantially augmented. That the augmentation was mediated by activation of SARM1 since no such change was seen in SARM1-KO mice infused with the Cx36-KD virus (Fig. 5e and Supplementary Fig. S5e-f).
Cx36-KD virus induced activation of SARM1 should lead to AxD and cell death. This was indeed the case. We monitored the cell number of premyelinating and myelinating oligodendrocytes by their marker, CC1 and the marker Tuj1 for neurons the brain slices by immunostaining. The results revealed that the Cx36-KD virus significantly diminished the CC1 signals in wildtype mice but not in SARM1-KO mice (Fig. 5f, green; Fig. 5g). The Tuj1 signals exhibited a similar trend, albeit less pronounced (Fig. 5f, orange). Correspondingly, behavioral tests, the open field and rotarod, showed demised performance of Cx36-KD virus-infused mice compared to those administered with the Scramble virus. SARM1-KO mice infused with Cx36-KD virus, however, were rescued and exhibited essentially normal performance (Fig. 5h and Supplementary Fig. S5g).