Multiplexed chemogenetics in astrocytes and motoneurons restore blood–spinal cord barrier in ALS

Chemogenetic motoneuron excitation and astrocyte GPCR-Gi signaling restore blood–spinal cord barrier, disrupted in four ALS mouse models, revealing its role in disease progression but not initiation.


Introduction
Disruption of the blood-spinal cord barrier (BSCB) is a recently appreciated feature of amyotrophic lateral sclerosis (ALS), a disease affecting upper and lower motoneurons (MNs) with a progressive course and invariably fatal outcome (Hardiman et al, 2017). In spinal cord samples from ALS patients, BSCB impairment has been detected in the form of plasma protein leakage and reduced expression of endothelial tight junction (TJ) proteins (Garbuzova-Davis et al, 2012). Likewise, extravasation of erythrocytes, immunoglobulins and plasma proteins; loss of endothelial TJ proteins (zonula occludens-1 [ZO-1], occludin, and claudin-5 [Zhong et al, 2008]); and decreased astrocytic end-feet and pericyte coverage (Garbuzova-Davis et al, 2007a, 2007b have been detected in SOD1 G37R , SOD1 G85R , and SOD1 G93A (Zhong et al, 2008;Winkler et al, 2014) as well as in SOD1 G93A rats (Garbuzova-Davis et al, 2007a, 2007b even at presymptomatic stages. Nevertheless, the driver(s) of BSCB disruption in ALS and its ultimate impact on disease progression are still debated. It has been hypothesized that damage of the BSCB may arise as a cell autonomous consequence of mutant SOD1 accumulation in endothelial cells: endothelial cells isolated from SOD1 G93A mice and immortalized mouse endothelial cells expressing the human G93A mutant SOD1 display reduced levels of claudin-5 (CLN-5) and decreased transendothelial resistance (Meister et al, 2015).
Finally, the impact of the BSCB disruption on disease onset and progression is not univocal. It has been suggested that the extravasation of blood-derived factors, inflammatory cells, erythrocytes, and hemoglobin degradation products in itself is able to trigger the degeneration of MN. Indeed, administration of activated protein C (APC [Zhong et al, 2009]) or of iron chelators (Winkler et al, 2014) to SOD1 G93A mice reduced the BSCB leakage and delayed disease progression. On the other hand, warfarin (an anticoagulant interfering with multiple vitamin K-dependent coagulation factors) aggravated BSCB disruption and worsened the degeneration of MN (Winkler et al, 2014). Nevertheless, none of these interventions are selective and may affect multiple disease cascades at once (e.g., reducing the transcription of the mutant SOD1 gene in the case of APC; Zhong et al, 2009).
Here, we have addressed the origin of the BSCB dysfunction in ALS and its weight on disease initiation and progression exploiting multiplexed chemogenetic strategies (involving both PSAM/PSEM 308 and DREADDs; Magnus et al, 2011;Roth, 2016) to enable interventions with high cell specificity and precise temporal control. We have demonstrated that BSCB disruption is a consequence of early, excitation-related MN dysfunction and that BSCB integrity can be restored (through the induction of Wnt proteins) by driving Gi signaling in astrocytes. Furthermore, we have used multiplexed chemogenetics to achieve a complete dissociation of BSCB integrity and disease burden, demonstrating how multiplexed chemogenetic can be used to untangle in vivo complex cellular interactions in BSCB disruption.

Results
Structural and functional disruption of the blood-spinal cord barrier is common to SOD1, FUS, TDP-43, and Tbk1 ALS mouse models We set out to investigate the extent and progression of BSCB disruption in four murine ALS models characterized by distinct genetic mutations and different degrees of phenotype severity and progression rate. In particular, we considered the high-copy SOD1 G93A line, together with the FUS ΔNLS/+ (Scekic-Zahirovic et al, 2017), the TDP-43 G298S (Wiesner et al, 2018), and Tbk1 heterozygous knockout (Tbk1 +/− [Brenner et al, 2019]) lines. In each ALS mouse strain, we investigated three time points at which critical pathological milestones were reached: (i) high-copy SOD1 G93A mice were studied at P20 (appearance of ER stress and misfolded SOD1 buildup; Saxena et al, 2013), P50 (denervation of highly vulnerable fast fatigable MN; Pun et al, 2006), and P80 (appearance of overt neurological signs; Boillée et al, 2006;Ouali Alami et al, 2018); (ii) FUS ΔNLS/+ mice were studied at P150 (before appearance of neurological signs and denervation), P270 (appearance of denervation and clasping), and P450 (more advanced neurological signs; Scekic-Zahirovic et al, 2017); (iii) TDP-43 G298S mice were studied at P150 (appearance of substantial neurological abnormalities; Wiesner et al, 2018), P360 (plateau of neurological dysfunction), and P510 (later stage in neurological dysfunction; subtle neurological abnormalities are present from the earliest time point but start worsening at around P150 with a plateau at around P250; Wiesner et al, 2018); and (iv) Tbk1 +/− mice were used as a reference line for ALS disease at P270 and P450, when they showed lack of motor symptoms, weight loss, or premature death (Brenner et al, 2019).
First, we verified that the expression of the human mutant SOD1 transgene and the expression of the mutant TDP-43 transgene were persistently overexpressed and comparable across all time points (Fig S1A-D). Likewise, we verified that in knock-in FUS ΔNLS/+ , the FUS protein was not overexpressed (as expected, Scekic-Zahirovic et al, 2016) but stayed stable over time (Fig S1E and F). Finally, TBK1 levels were steadily reduced in Tbk1 +/− mice compared with WT mice ( Fig  S1G and H).
We further explored the possibility that TJ loss may be due to the focal degeneration of endothelial cells (in the so-called string vessel formation, defined as a collapsed basement membrane without the endothelium and with no function in circulation; Forsberg et al, 2018). We immunostained spinal cord sections for COL-IV + together with CD31/PECAM-1, a membrane and cytoplasmic protein abundantly and constitutively expressed on endothelial cells (Feng et al, 2004;Caligiuri, 2019). Across all ALS lines and all time points, we observed an almost complete (>90%) coverage of COL-IV + with CD31/PECAM-1 + immunostaining, confirming that every COL-IV + vessel was indeed lined with endothelial cells (Fig S3).
Consequently, these data show not only that structural and functional BSCB disruption is a generalized phenomenon discovered in all four ALS mouse models but also that BSCB breakdown is already detectable at stages in which there is no MN loss (P20 in SOD1 G93A , P150 in FUS ΔNLS/+ , and P270 and P450 in Tbk1 +/− mice).

Chemogenetic inactivation of MN firing enhances BSCB disruption, whereas stimulation of MN excitation restores BSCB integrity
We therefore established that BSCB disruption is an early event both shared by ALS mouse lines expressing different pathogenic mutations and appreciable before MN loss (in the SOD1 G93A and FUS ΔNLS/+ mice). We then investigated the pathogenic mechanisms potentially responsible for initiating the BSCB disruption. Loss of excitability in vulnerable MN is among the earliest manifestations of disease (Martinez-Silva et al, 2018), and enhancement of MN excitation has a direct beneficial effect on the burden of misfolded proteins, such as misfolded SOD1, as well as on ER stress and autophagy overload (Saxena et al, 2013;Bą czyk et al, 2020). We therefore set out to investigate if early changes in MN excitability may be causally related to the disruption of the BSCB. For this purpose, we exploited an engineered ion channel with orthogonal pharmacology (PSAM/PSEM; [Magnus et al, 2011]), to either stimulate (cation-permeable PSAM: actPSAM) or inactivate (anionpermeable PSAM: inhPSAM) neuronal activity upon administration of the pharmacologically selective synthetic ligand (PSEM 308 hydrochloride [PSEM 308 ]). Intraspinal injections with AAV9 vector (encoding either the actPSAM or the inhPSAM in double-inverted orientation under the human synapsin promoter: hSyn::DIO-actPSAM or hSyn::DIO-inhPSAM.EGFP) were performed in P20 SOD1 G93A /ChAT-Cre double-tg mice ( Fig 7A). Starting from P28, when a robust MN expression of the PSAM was observed, the mice were daily treated for 7 d with PSEM 308 or with vehicle ( Fig 7B). We verified from the two previous images.   showing that actPSAM activation resulted in a significant up-regulation of the activity-dependent gene product DREAM/KChIP3 in the nuclei of infected MNs, compared with the uninfected contralateral MNs (average fluorescence intensity per nucleus: 2,073 ± 475 a.u. versus 1,402 ± 104, respectively; P < 0.0001; Fig 6A and B). Likewise, S133phosphorylated CREB ([pCREB] Wu et al, 2001) was increased in the actPSAM-infected MNs compared with the contralateral uninfected MNs (1,694 ± 723 a.u. versus 1,287 ± 330 a.u., respectively; P < 0.0001; Fig 6C and D). Conversely, inhPSAM caused a significant decrease in both activity markers: DREAM (1,043 ± 103 a.u.; inhPSAM versus uninfected; P < 0.01; inhPSAM versus actPSAM; P < 0.0001, Fig 6A  The experiment proved that intraspinal injection per se did not produce any permanent disruption of the BSCB. Intriguingly, chemogenetic activation of MN firing, by actPSAM/PSEM 308 , caused a significant decrease in the BSCB disruption compared with the uninfected ventral horn, with breaks in CLN-5 ribbon reduced up to 59.3% ± 13.1% of contralateral (P < 0.0001; Fig 7D, G, and J). By contrast, the inactivation of MN firing by inhPSAM/PSEM 308 resulted in increased fragmentation of the CLN-5 ribbon in the infected ventral horn (129.8% ± 12.7% of contralateral), compared with the noninfected contralateral ventral horn (P < 0.001; Fig 7E, H, and K; actPSAM/ PSEM 308 versus inhPSAM/PSEM 308 : P < 0.0001; Fig 7J and K). These data suggest that BSCB disruption is dependent upon the early excitation-associated dysfunction of MN.
Chemogenetic activation of astrocytic Gi and Gq signaling is sufficient to restore BSCB integrity The application of chemogenetic PSAM/PSEM 308 to control MN firing in an ALS mouse model (SOD1 G93A ) established a causal link between MN activity and stability of the BSCB in ALS. Two sets of considerations led us to hypothesize that astrocytes may provide a link between neuronal firing and the BSCB in an ALS disease condition: (i) astrocytes sense and respond to increased synaptic activity and neuronal firing by detecting the release of (among others) ATP, glutamate, and acetylcholine Kofuji & Araque, 2020) as well as K + ions (Simard & Nedergaard, 2004); (ii) astrocytes are involved in the control of BSCB differentiation and integrity (Abbott et al, 2006;Liebner et al, 2018) by releasing GDNF (Igarashi et al, 1999), angiopoietin-1, and bFGF (Lee et al, 2003). Considering that neuronal activity is monitored by astrocytes via different heteromeric G protein-coupled receptor (GPCR) families and signaling through Gα proteins (Gs, Gi, or Gq; Kofuji & Araque, 2020), we hypothesize that controlling the Gs, Gi, or Gq signaling cascade through designer receptors (DREADDS, coupled to Gs [D(Gs)], Gi [D(Gi)], or Gq [D(Gq)]), we could provoke astrocyte responses related to increased neuronal activity and possibly restore the integrity of the BSCB without intervention on MN themselves. We therefore injected AAV8 encoding GFP alone or citrine-tagged DREADDs coupled to Gs, Gi, or Gq under the control of the GFAP mini promoter (Fig 8A and B) in the spinal cord of P20 SOD1 G93A /ChAT-Cre mice (although the expression of Cre is not necessary to target astrocytes, we used the same line as the PSAM/ PSEM 308 experiments to maintain a comparable background). Clozapine-N-Oxide (CNO, or vehicle) was administered from P28 until P35 (Fig 8A). The integrity of the CLN-5 ribbon was then evaluated by comparing the injected ventral horn of the spinal cord to the contralateral uninjected ventral horn as an internal control (subject to the same treatments and pharmacology but not receiving the virus). Under these conditions, we verified that DREADDs or GFP was expressed in 65-90% of GFAP + cells in the ventral horn of the spinal cord (insert in Fig 8B). When compared with the contralateral ventral horn, the GFP alone in astrocytes (under CNO treatment) did not affect the extent of BSCB breakdown (CLN-5 breaks were 95.8% ± 21.8% of the uninfected contralateral ventral horn; Fig 8C, G, and K). Likewise, expression of DREADDs followed by vehicle treatment did not result in significant changes in BSCB breakdown in injected versus non-injected sides. Taken together, these data confirm that neither CNO treatment nor AAV injection per se affects BSCB integrity. However, when we expressed D(Gi) in astrocytes and administered CNO, the injected side displayed a significant restoration of CLN-5 distribution compared with the contralateral, non-injected side (52.5% ± 12.7% of contralateral; P < 0.001; Fig 8D, H, and L). A similar effect was also detected when D(Gq) was activated (49.2% ± 1.3% of the contralateral side; P < 0.0001; Fig  8E, I, and M) but not in the case of D(Gs) (103.1% ± 27.8% of the contralateral uninfected side; Fig 8F, J, and N).
We further evaluated the effect of DREADD signaling activation on the astrocytic end-feet coverage of spinal cord capillaries ( Fig  9A-G), an additional parameter to evaluate BSCB integrity. Compared with GFP-expressing astrocytes (Fig 9A-C  resulted in a significant increase in vascular end-feed coverage (51.2% ± 11.8% in D(Gi) versus 30.7% ± 12.0% in GFP-expressing astrocytes; P < 0.01; Fig 9B and C). Likewise, D(Gq) also increased end-feet coverage (52.9 ± 13.0; P < 0.01 versus GFP; Fig 9B and C), whereas activation of DREADD-(Gs) did not significantly affect the coverage of microvessels by astrocyte end feed (40.5% ± 8.4%; P < 0.1; Fig 9B and C). The effect on astrocyte end feet was additionally confirmed by monitoring the levels of aquaporin 4 (AQP4 [Bataveljic et al, 2014]) surrounding the microvessels, located in the ventral horn of the spinal cord. In animals injected with D(Gi) and treated with CNO, the AQP4 + coverage of microvessels was strongly increased compared with the contralateral uninjected ventral horn of the spinal cord (25.7% ± 6.8% versus 12.6% ± 4.8%, respectively, P < 0.0001; Fig 9D and E). Moreover, we also verified that AQP4 + coverage of vessels also responded to chemogenetic control of MN excitability. Indeed, MN inactivation by inhPSAM, resulted in the significant decrease in AQP4 + coverage of vessels compared with the contralateral ventral horn (7.7% ± 3.1% versus 12.1% ± 4.1%, respectively; P < 0.001; Fig 9F and G). Taken together, these data show that activation of Gi-and Gqcoupled signaling cascades in astrocytes successfully restores BSCB integrity in presymptomatic SOD1 G93A mice.

Activation of astrocytic D(Gq) but not D(Gi) ameliorates MN disease markers
Next, we tested if the restoration of BSCB integrity by D(Gi) and D(Gq) correlates with beneficial or detrimental effects on the MN burden of established disease markers (Fig 10A). We considered the accumulation of the misfolded protein SOD1 (using the conformation-specific antibody B8H10; Bosco et al, 2010), ER stress levels (KDEL; Saxena et al, 2013), and the burden of p62 + inclusions (Rudnick et al, 2017;Martinez-Silva et al, 2018;Ouali Alami et al, 2018). The levels of misfolded SOD1 in MN (magenta arrows in Fig 10B), proximal to D(Gq)-activated astrocytes, were significantly lower than those in contralateral MN (orange arrows in Fig 10B [935 ± 346 versus 1,303 ± 342 a.u., respectively, P < 0.0001; Fig 10C]). Likewise, activation of astrocytic D(Gq) decreased the number of p62 + inclusions (1.1% ± 1.5% versus 4.3% ± 4.7% aggregate burden/total cell body, P < 0.0001; Fig 10D and E) and ER stress (1,495 ± 459 versus 1,621 ± 358 a.u., respectively; P < 01; Fig 10F and G) in nearby MNs (magenta arrows), when compared with the contralateral noninfected ventral horn (orange arrows). Surprisingly, activation of D(Gi) in astrocytes did not affect the accumulation of misfolded SOD1 in nearby MN (743 ± 222 versus 730 ± 130 a.u., respectively, P = 0.54; Fig 10H and I) nor p62 + aggregates (4.3% ± 37.2% versus 4.7% ± 36.6% of aggregate burden/ total cell body in the infected versus noninfected ventral horn; P = 0.60; Fig 10J and K) or KDEL levels (1,343 ± 433 versus 1,437 ± 343 a.u., infected versus noninfected ventral horn; P = 0.08; Fig 10L and M). We further extended the investigation of D(Gi) by considering the accumulation of the autophagy marker LC3A. In this case too, D(Gi) did not improve the accumulation of LC3A buildup in MNs (1,475 ± 232 versus 1,431 ± 296 a.u., respectively; Fig 10N and O). Whereas activation of astrocytic D(Gq) drives a generalized amelioration of the disease pathways (MN disease markers and BSCB integrity), stimulation of D(Gi) in astrocytes produces a dissociation between restored BSCB integrity and unmodified disease markers in MN. The D(Gi) findings imply that at this stage, BSCB impairment has limited effect on disease pathways because the BSCB can be restored without impacting ALS pathobiochemistry.

Restoration of BSCB by astrocytic D(Gi) signaling is independent of MN firing
We then focused on D(Gi), considering that its activation dissociates the effect on BSCB integrality from MN disease markers, which could potentially shed light on the role of BSCB disruption on the pathogenesis of MN disease. We therefore investigated the mechanism involved in astrocytic D(Gi)-mediated restoration of BSCB. Because increasing MN firing was sufficient to improve BSCB integrity, we explored whether astrocytic Gi would prove beneficial by modulating MN activity, for example, by inducing the release of excitatory neurotransmitters . In this case, we reasoned that any beneficial effect of D(Gi) on BSCB would be significantly decreased by the concomitant inactivation of MN firing. We arranged a multiplexed chemogenetic strategy to inactivate MN firing using the inhPSAM/PSEM 308 , while simultaneously triggering Gi signaling in astrocytes by D(Gi) (Fig 11A and B). To this end, intraspinal injections were performed in SOD1 G93A /ChAT-Cre mice with AAV9 encoding for hSyn::DIO-inhPSAM (expression restricted to MN) and with AAV8 encoding GFAP::D(Gi) (expression restricted to astrocytes). After 10 d, we administered to the animals both CNO and PSEM 308 (or vehicle + PSEM 308 ) for a further 7 d (Fig 11B). Inactivation of MN alone resulted in the anticipated increased disruption of the BSCB (129.8% ± 8.7% of the contralateral noninfected ventral horn; P < 0.01; Fig 11C, E, and G). However, the activation of D(Gi) in the presence of MN inactivation still resulted in a substantial improvement of BSCB integrity (68.6% ± 10.7% of the contralateral uninfected ventral horn; P < 0.01; Fig 11C, E, and G). Nevertheless, MN firing inactivation by inhPSAM/PSEM 308 , even when co-applied with astrocytic D(Gi) activation, caused increased levels of LC3A (1,840 ± 396 versus 1,693 ± 395 a.u., respectively, in the COL-IV + vessels (green) in the infected ventral horn of the spinal cord compared with the contralateral uninfected ventral horn.  . Representative pictures of aquaporin-4 (green) expression localized in the astrocytic end-feet enveloping microvessels (red) in the ventral horn of the co-injected ventral horns versus contralateral uninjected horns; P < 0.0001; Fig 11H) and a higher burden of p62 + inclusions (4.6% ± 3.1% versus 3.2% ± .1.8% of aggregate burden/total cell body, coinfected versus contralateral; P < 0.1; Fig 11I), confirming the effective target engagement of inhPSAM/PSEM 308 (Saxena et al, 2013) and further demonstrating that BSCB integrity, and disease marker burden in MN can be dissociated. We further explored whether D(Gi) activation would confer an additional beneficial effect when applied together with increased MN firing (i.e., if the two beneficial effects are additive or not). We compared the CLN-5 ribbon integrity in spinal cord sections in which both MN firing (actPSAM/PSEM 308 ) and astrocytic D(Gi) were activated. Notably, single chemogenetic and multiplexed chemogenetic samples displayed the same degree of BSCB restoration, (actPSAM/PSEM 308 + D(Gi)/CNO: 60.9% ± 6.7% of contralateral noninfected versus 59.2% ± 13.1% in actPSAM/PSEM 308 + D(Gi)/veh; P > 0.05; Fig 11D, F, and J). Driving MN excitation by actPSAM/PSEM 308 therefore not only mimicked but also truly occluded the effect of astrocytic Gi signaling.
Taken together, the double-chemogenetic experiments imply that the consequences of D(Gi) on BSCB are independent of MN firing, although they mimic the effects of MN activity stimulation, suggesting that astrocytic Gi signaling may be located downstream of MN excitation.

Astrocytic Wnt7a and Wnt5a levels are enhanced by Gi signaling in astrocytes and suppressed by MN inactivation, in correlation with BSCB modifications
We sought to identify the mediator(s) involved in the sealing of the BSCB induced by the activation of Gi signaling in astrocytes. We focused on the Wnt family because these mediators are necessary in the establishment of the blood-brain-barrier (BBB) during development (Liebner et al, 2008), as well as in its maintenance in adulthood (Tran et al, 2016;LeBlanc et al, 2019). First, we explored if the blockade of Wnt signaling modified the integrity of the BSCB in WT and SOD1 G93A mice. The animals were treated for 14 d either with vehicle or with the porcupine (PORCN) inhibitor (which blocks Wnt acylation and release) C59 (Proffitt et al, 2013;Torres et al, 2019), starting from day P20. The administration of the Wnt antagonist C59 caused a small disruption of the BSCB in the WT mice (124.7% ± 16.4% of WT control mice treated with vehicle) and a substantial worsening of the BSCB integrity in the SOD1 G93A mice (157.2% ± 17.7% of WT veh; P < 0.0001; Fig 12A and B), indicating that Wnt proteins are involved in balancing the BSCB integrity in the ALS mice.
Next, we focused on Wnt5a, the most abundant Wnt family member in the spinal cord (Ouali Alami et al, 2018), recognised as sufficient for inducing a BBB-like phenotype in endothelial cells (Artus et al, 2014) and on Wnt7a, a critical mediator in the establishment and maintenance of the BBB phenotype (Cho et al, 2017;Wang et al, 2018). We exploited single-molecule mRNA in situ hybridization (coupled to immunostaining for the identification of cellular subpopulations) to quantify Wnt5a and Wnt7a in spinal cord astrocytes (identified by GFAP staining). Both Wnt5a and Wnt7a mRNA molecules were readily detected in GFAP + astrocytes in the WT and SOD1 G93A spinal cord sections (Fig 12C-F). Notably, both astrocytic Wnt5a mRNA molecules (58.6% ± 12.3% of WT; P < 0.01; Fig  12C and D) and Wnt7a mRNA molecules (28.6% ± 11.9% of WT; P < 0.0001; Fig 12E and F) were significantly decreased in the SOD1 G93A mice compared with WT controls.
We then explored if Wnt5a and Wnt7a levels could be modulated by chemogenetic manipulations in correlation with the integrity of the BSCB. We determined that chemogenetic inactivation of MN firing by inhPSAM/PSEM 308 caused a significant additional decrease in Wnt5a mRNA (40.5% ± 13.1% of contralateral; P < 0.0001; Fig 12G  and H) and in the Wnt7a mRNA molecules (71.9% ± 11.6% of the uninfected contralateral horn; P < 0.01; Fig 12I and J) in the astrocytes of the infected ventral horn compared with the contralateral side. These data suggest that disruption of the BSCB driven by MN inactivation is associated with the reduced expression of Wnt5a and Wnt7a in astrocytes. We tested this hypothesis by assessing the impact of astrocytic Wnt5a and Wnt7a levels on D(Gi) activation in astrocytes. We obtained a strong increase in Wnt5a (180.8% ± 33.3% of contralateral; P < 0.01; Fig 12K and L) and Wnt7a mRNA molecules (219.4% ± 6.4%; P < 0.01; Fig 12M and N) in the infected astrocytes compared with the uninfected astrocytes in the contralateral side. Wnt5a and Wnt7a levels were therefore highly correlated with the integrity of BSCB during the pharmacological inhibition of Wnt signaling, as well as with an increase in MN firing and stimulation of astrocytic Gi signaling.

Activation of astrocytic Gi signaling at later stages ameliorates both BSCB disruption and disease burden
Because ALS pathogenic processes are heterogeneous and evolve over time (Ouali Alami et al, 2018), we investigated whether prolonged D(Gi) activation would enable the restoration of the BSCB in correspondence to the first wave of MN denervation (taking place at about P50; [Pun et al, 2006]). SOD1 G93A mice were injected with AAV8(GFAP::D(Gi)-Citrine) at P20 and treated with a single injection CNO daily for 20 d, from P30 until P50 (Fig 13A). We observed that prolonged activation of astrocytic D(Gi) resulted in a significant decrease in the breakdown of the BSCB (CLN-5 breaks length, 59.9% ± 8.0% in D(Gi) versus uninfected contralateral controls; P < 0.001; Fig 13B and D). Surprisingly, the prolonged activation of astrocytic D ( Fig 13E and G). At the same time, the amount of BSCB restoration remained substantial even upon simultaneous MN inactivation by inhPSAM (CLN-5 breaks length, 72.7% ± 8% of uninfected contralateral side; P < 0.0001; Fig 13H-L), confirming the independence of the astrocytic D(Gi) from MN excitability effects.
In conclusion, although BSCB restoration, driven by Gi signaling in astrocytes at early stages of the disease, was not associated with any effect on disease markers, it could be linked to neuroprotective effects at later phases of ALS pathology.

Discussion
In the present work, we have showed that (i) BSCB structural and functional disruption is detected in four ALS mouse lines with unrelated mutations and different degrees of MN loss; (ii) the disruption of the BSCB takes place before any MN loss (in at least two lines), and it is restored by enhancing MN excitability; and (iii) the activation of Gi signaling in astrocytes reverts the disruption of the BSCB by increasing Wnt5a/Wnt7a expression but without impacting the burden of disease markers in MN and independently of MN firing. Overall, these data suggest first that in healthy conditions, MN excitability controls BSCB integrity through astrocytes and second that in ALS, the loss of MN excitability and firing leads (possibly through the decrease in signaling driven by the Gicoupled receptor) to the down-regulation of Wnt5a/7a in astrocytes, which in turn provokes the opening of the BSCB. Whereas at the very initial stages, the restoration of BSCB integrity does not affect MN pathobiochemistry, the opposite is true later on: the disruption of the BSCB is an event secondary to MN dysfunction, which then feeds back on MN and aggravates the ongoing pathogenic process.
Although the disruption of the BSCB in ALS has been previously reported (Garbuzova-Davis et al, 2007a, 2007bZhong et al, 2008), the nature of the primum movens has remained debated: Is BSCB impairment driven by MN dysfunction or is it a MN-independent event? Some (controversial) evidence has pointed toward an MNindependent origin: Endothelial cells expressing mutant SOD1 in vitro display a cell-autonomous disruption of TJs because of the misfolded protein itself (Meister et al, 2015). At the same time, endothelial cell-selective excision of the mutant SOD1 transgene does not prevent the breakdown of the BSCB nor affect the survival of the transgenic mice (Zhong et al, 2009). The loss of pericytes observed in spinal cord samples from ALS patients has been hypothesized to be at the origin of BSCB disruption (Winkler et al, 2013). However, increased PDGF-C signaling appears to be associated with BSCB disruption in ALS (Liebner et al, 2008), even if PDGFR-α signaling is actually necessary for the survival of these cells (Kisler et al, 2017). Recently, microglial activation, alone or together with peripheral immune cells contribution, has been hypothesized to be involved in the opening of the BSCB (Puentes et al, 2016;Frakes et al, 2017;Epperly et al, 2019). We have demonstrated that chemogenetic control of MN firing can bidirectionally modulate BSCB integrity: Increased firing reduces BSCB disruption, whereas neuronal inactivation worsens it. This pattern corresponds to what has been observed for other disease markers such as misfolded SOD1 accumulation, ER stress, and unfolded protein response (Saxena et al, 2013;Bą czyk et al, 2020). Even if other cell types play contributing roles, the impairment of the BSCB is thus the manifestation of a pathogenic process, primarily originating in MN. Notably, BSCB impairment is detected in ALS mouse models before any MN loss (such as in the FUS ΔNLS and SOD1 G93A ) but at stages when hypoexcitability of vulnerable MN is already detectable (Martinez-Silva et al, 2018). Thereafter, the disruption of the BSCB may be a consequence of early MN dysfunction.
How is this dysfunction translated into a vascular phenotype? Considering that astrocytes are sensitive to neuronal activity (through several glutamatergic, GABAergic, cholinergic, and purinergic GPCR; Kofuji & Araque, 2020) and, at the same time, they are key organizers of the BSCB (Abbott et al, 2006), these cells become prime candidates for investigation. We exploited DREADDs with astrocyte-selective promoters to obtain a direct cell-and timespecific modulation of GPCR signaling (Orr et al, 2015;Bang et al, 2016;Adamsky et al, 2018;Jones et al, 2018;Nagai et al, 2019). We demonstrated that D(Gq) and D(Gi) can induce the restoration of BSCB integrity, although D(Gq) itself appears to be beneficial toward all disease readouts, suggesting that it may restore the BSCB by ameliorating MN firing, possibly upon triggering gliotransmitter release . Critically, D(Gi) signaling acts directly on the BSCB, independently of MN firing, as we have shown in our multiplexed chemogenetic strategy.
How does astrocytic D(Gi) induce the restoration of the BSCB? We elected to investigate the role of Wnt proteins, in particular Wnt5a and Wnt7a. Wnt family members are involved in the establishment of the BBB during development (Liebner et al, 2008;Stenman et al, 2008) and contribute to its maintenance in the adult brain (Artus   Tran et al, 2016). Low Wnt/β-catenin signaling actually characterizes the site in the brain with the lowest integrity of the BBB and the highest permeability (Benz et al, 2019). Among Wnt family members, Wnt7a appears to be critical during development (Daneman et al, 2009) and maintains the integrity of the BBB through Gpr124, during a stroke or in glioblastoma (Chang et al, 2017), as well as in homeostatic conditions. Nevertheless, Wnt5a is sufficient to induce a BBB phenotype in endothelial cells in vitro (Artus et al, 2014), and it is among the most strongly expressed Wnt members in astrocytes (Ouali Alami et al, 2018). We have demonstrated that Wnt5 expression in astrocytes is reduced early on in SOD1 G93A mice, and the suppression of Wnt signaling by the porcupine inhibitor C59 further worsens BSCB disruption. Moreover, the astrocytic Wnt5a mRNA and Wnt7a mRNA levels bidirectionally correlate with the integrity of the BSCB, being down-regulated upon inactivation of MN firing (worsening of the BSCB integrity) and up-regulated by D(Gi) activation in astrocytes (ameliorating the BSCB integrity). During increased availability of Wnt5a and Wnt7a, it is believed that the activation of the β-catenin signaling cascade in endothelial cells is a key event in inducing and maintaining BSCB specializations. Canonical Wnt signaling through β-catenin is sufficient to induce a BBB-like phenotype in cultured endothelial cells (Laksitorini et al, 2019), and it is critically involved in establishing the BBB phenotype in vivo (Liebner et al, 2008;Chang et al, 2017;Cho et al, 2017;Wang et al, 2018), whereas reduced Wnt/β-catenin signaling spatially corresponds to sites of high BBB permeability in the brain (Benz et al, 2019). Moreover, noncanonical Wnt signaling may contribute to the establishment of the BBB phenotype (Pinzón-Daza et al, 2014). Sox17 has been identified as a major target of the β-catenin pathway in endothelial cells (Corada et al, 2019), together with Lef1 and Ets1 (Roudnicky et al, 2020). These Wnt-regulated transcription factors up-regulate the transcription of several junctional proteins, including VE-cadherin and claudin-5 (Roudnicky et al, 2020), possibly involved in the repair of the BSCB. Furthermore, claudin-5 is dynamically internalized and stored in intracellular vesicles upon inflammatory stimuli and trafficked back to the cell surface when the inflammatory signaling is subsided (Stamatovic et al, 2003(Stamatovic et al, , 2006(Stamatovic et al, , 2009. It is therefore conceivable that transcriptional mechanisms and trafficking may be responsible for Wnt5a/7a-driven BSCB restoration. In principle, it is possible that additional mediators may contribute to the D(Gi) effect (e.g., GDNF; Igarashi et al, 1999). One could also hypothesize that the up-regulation of Wnt5a/7a induced by D(Gi) may affect other nonneuronal cells (e.g., microglia) and only indirectly play a role in the integrity of the BSCB. Although this model is theoretically possible and worth further investigation, the direct effects of Wnt5a/7a on BSCB integrity are well established (Liebner et al, 2008;Stenman et al, 2008;Artus et al, 2014;Tran et al, 2016;Cho et al, 2017;Wang et al, 2018;Benz et al, 2019;Laksitorini et al, 2019;LeBlanc et al, 2019). Because microglia do respond to astrocytic Wnt (Ouali Alami et al, 2018), Wnt signaling is thought to cause microglial proliferation and substantial pro-inflammatory activation (Halleskog et al., 2012(Halleskog et al., , 2011, although the ultimate outcome may depend on the context (Halleskog and Schulte, 2013). Putative indirect actions of astrocytic D(Gi) remain to be further investigated.
In our current interpretative model (Fig S5), MN activity is hypothesized to be linked to BSCB integrity via astrocytes and astrocyte-secreted Wnt5a/7a. Nevertheless, to date, virtually nothing is known about the MN-originated signals that modulate Wnt5a/7a in adult astrocytes. However, one can speculate about the role of astrocytic GPCRs involved in monitoring neuronal and synaptic activities (e.g., GABA, glutamate, ATP, and acetylcholine receptors are expressed by astrocytes; Kofuji & Araque, 2020). Insufficient engagement of these receptors due to MN hypoexcitability and reduced synaptic excitation (Martinez-Silva et al, 2018;Bą czyk et al, 2020) may also decrease Wnt5a/7a in astrocytes. One could further speculate that MN-initiated signaling might converge on NF-kB because overactivation of IKK-β in astrocytes is sufficient to increase Wnt5a expression (Ouali Alami et al, 2018).
The ability to control BSCB integrity in vivo enables us to address a critical issue in the vascular biology of ALS: Is the disruption of the BSCB detrimental to the MN disease process? Previous evidence suggests that closure of the BSCB by APC administration is associated with better outcomes; however, this effect was linked in reality to the decrease in the expression of the mutant SOD1 gene itself induced by APC (Zhong et al, 2009). Likewise, the worsening of disease progression observed upon warfarin administration has been attributed to the substantial disruption of the BSCB, the extravasation of erythrocytes, and iron-triggered radical oxygen toxicity (Winkler et al, 2014). Nevertheless, this interpretation is complicated by the unclear target selectivity of warfarin itself and by the lack of an identified cellular target. There is further proof that even the substantial disruption of the BSCB that follows NF-κB activation in astrocytes is not necessarily associated with a worsening of the disease progression (Ouali Alami et al, 2018). Furthermore, mutant SOD1 mice lacking aquaporin-4 display an intact BSCB immunostaining for the multiplexed chemogenetic experiments in which MN firing inhibition via inhPSAM/PSEM 308 (infected MNs in green) combined with astrocytic Gi activation diminished the breaks (yellow lines) in    but do not have a better disease course and survival (Watanabe-Matsumoto et al, 2018). Our data show that the activation of D(Gi) in astrocytes restores BSCB integrity without ameliorating the burden of multiple disease markers in MN (effectively dissociating the BSCB from other disease manifestations). This suggests that at the initial stages of the disease, BSCB impairment does not necessarily carry pathogenic consequences.
At the same time, our findings are compatible with a possible pathogenic role of BSCB impairment later in disease progression because at this stage, Gi signaling in astrocytes restores the BSCB (once again independently of MN firing), while it reduces the burden of disease markers. However, some limitations apply to later-stage results: Gi signaling activates a number of biological processes in astrocytes, whereas the role of astrocytes may change over time (Ouali Alami et al, 2018). Although it is not possible to draw an unequivocal causal link between BSCB restoration and disease burden, it is still plausible that at this stage, BSCB restoration might be the consequence of the decrease in disease burden.
In conclusion, we demonstrated that BSCB impairment originates as a consequence of early MN dysfunction and can be dissociated from MN disease burden through the selective activation of Gi signaling in astrocytes and the consequent induction of Wnt proteins (in particular Wnt5a/7a). This discovery implies that BSCB disruption does not have a pathogenic role at the early stages of the disease. However, prolonged Gi signaling in astrocytes beneficially affects BSCB function at later stages of the disease by decreasing or delaying MN disease burden.
All experiments were carried out on male mice because of the known difference in disease progression rate between male and female mice and of the less predictable time course in female mice, starting from the age of P20. Unless otherwise specified, animals were kept in groups of 3-4 mice in a 12-h light-dark cycle and were given ad libitum access to food and water. Mice were checked daily for the appearance of symptoms. Based on power calculations (α = 80% with 5% significance two-sided) for each experimental group or time points, 6-8 animals were processed and analyzed.

Intraspinal injection of AAV
Intraspinal injection of AAV was performed as previously reported (Saxena et al, 2013). Briefly, the spinal cord was accessed by a dorsal laminectomy performed at the T11-T13 level. Injection (1 μl) was performed using a pulled glass capillary coupled to a Picospritzer-III apparatus. Using the central dorsal artery as reference, injections were performed at the coordinates y = +0.30; z = −0.45. Further details are reported in Supplemental Data 1.

Chemogenetics agonist administration
CNO was purchased from Tocris and was administered by i.p. injection at a dose of 5 mg/kg once daily, starting 10 d after the intraspinal injection (unless otherwise specified) for 7 consecutive days. The effector molecule PSEM 308 was custom synthesized by Apex Scientific and was administered daily at a dose of 5 mg/kg in saline, starting from 10 d after viral injection for 7 consecutive days; because PSEM 308 is poorly soluble in cold saline, stock solutions were pre-warmed at 40°C for 1 h before injection and used at RT.

Western blot
For the preparation of whole spinal cord protein extract, the lumbar spinal cord was dissected after mouse euthanasia by cervical dislocation and snap frozen on dry ice as previously reported (Olde Heuvel et al, 2019). Briefly, lumbar spinal cord tissue was dissected and homogenized in a complete radioimmunoprecipitation assay (RIPA) buffer, containing protease and phosphatase inhibitors. Samples were then subjected to SDS/PAGE gel electrophoresis, transferred to a polyvinylidene difluoride (PVDF) or nitrocellulose membrane, and subsequently incubated overnight at 4°C with the following primary antibodies: total anti-TDP 43 rabbit (10782-2-AP; 1:2,000 buffered in 1% BSA in PBS containing 0.05% Tween 20; Proteintech), anti-FUS rabbit (1:2,000 buffered in 1% BSA in PBS containing 0.05% Tween 20; Bethyl Laboratories), and anti-SOD1 rabbit (Prestige 001401; 1:1,000 buffered in 1% BSA; in PBS containing 0.05% Tween 20; Sigma-Aldrich). After washing steps in PBS enriched with 0.05% Tween 20, the membranes (nitrocellulose membrane for SOD1 and PVDF for TDP43 and FUS) were subsequently incubated for 1 h at RT with goat anti-rabbit IgG-HRP-conjugated secondary antibody (1: 10,000 in PBS-Tween 20; Bio-Rad). After appropriate washing steps, the membranes were further treated with Western ECL-immunodetection buffer (Bio-Rad) and acquired using Image Lab 5.0. ThermoScientific PageRuler Plus Prestained was used as a protein ladder. Samples were corrected for background, and densitometry analysis was performed using Image Lab software5.0. To control variability in loading individual samples, the signal intensity of protein bands was normalized to housekeeping GAPDH protein.

Histology and immunostaining
The spinal cord was prepared for histology and immunostaining as previously reported (Saxena et al, 2013;Ouali Alami et al, 2018). Briefly, after perfusion, fixation with 4% PFA, and cryoprotection in 30% sucrose, 40-μm-thick sections were subject to antigen retrieval (whenever indicated) and immunostained using a free-floating protocol. Details of the immunostaining procedure are reported in Supplemental Data 1, and a list of the antibodies and other reagents used is reported in Table S1.

Claudin-5 breaks quantification
The loss of CLN-5 ribbon is defined by a lack of CLN-5 immunostaining under specific threshold by taking as reference the capillary wall stained with collagen-IV. The length breaks are measured and traced (in yellow) with ImageJ software. The quantification is performed by relating the total length of the single breaks and the total length of the vessels, outlined by collagen-IV, and expressed as % of breaks on the total vessels. Images are acquired as single tile scans, covering the ventral horn of the spinal cord, with specific Z-stack at the confocal microscope and analyzed as collapsed pictures.

Single-molecule in situ mRNA hybridization
Detection of mRNA in situ together with co-immunostaining was performed as previously reported (Olde Heuvel et al, 2019) and according to the manufacturer's recommendation (Acd Bio). Details of the procedure are reported in Supplemental Data 1, and details of the probes used are reported in Table S1.

Confocal imaging and image analysis
Confocal images were acquired as previously reported (Ouali Alami et al, 2018). Details of image acquisition are reported in Supplemental Data 1.
For the quantification of the structural disruption of the BSCB, we considered a ROI in the ventral spinal cord. We traced the total length of vessels, covered by collagen-IV staining, in the ROI and the length of vascular segments in which the claudin-5 ribbon appeared disrupted or fragmented ("gaps" or "breaks," depicted in yellow lines) and computed the ratio between the cumulative gap length and the total length of the vessels, as previously reported (Ouali Alami et al, 2018). breaks length. (E) Representative pictures showing MN markers p62 and LC3A immunofluorescence staining (gray) in infected and uninfected contralateral ventral horn of SOD1 G93A /ChAT-cre spinal cord sections. MNs are detected by VAChT immunostaining (red) and infected astrocytes by GFP (green). (F, G) Quantification of (F) p62 aggregates per cell body and (G) LC3A intensity in MN surrounding infected astrocytes and in MN located in the uninfected contralateral ventral horn. The quantifications are represented by the box-and-whisker plot; 10-90 percentile is considered. Data are from N = 3 mice. (I) Experimental design for multiplexed chemogenetic injection of AAV9 encoding inhPSAM + D(Gi) in SOD1 G93A /ChAT-cre mice, at P20 and treated with respective ligands PSEM 308 and CNO from P30 until P50. (H) Expression pattern of MNs activated by inhPSAM (green) and astrocytes activated by D(Gi). Dotted lines delineate the contour of the gray and white matter in the ventral horns of the spinal cord. The insert highlights infected MNs (green) surrounded by infected astrocytes (green). Scale bar: 50 μm. (J) Representative panel displaying the effect of the prolonged and late activation of multiplexed chemogenetic inhPSAM/PSEM 308 + D(Gi)/CNO experiments on the BSCB grade of disruption in SOD1 G93A /ChAT-cre mice. MNs in the noninfected horn are identified by VAChT immunostaining (blue), whereas infected MNs and astrocytes are stained with GFP (green). CLN-5 breaks are depicted in yellow lines along the ribbon (red). (K) High magnification of CLN-5 organization (white) along COL-IV + vessels (red). Yellow arrows indicate CLN-5 interruptions along the ribbon in the noninfected ventral horn of SOD1 G93A /ChAT-cre mice subjected to the late and prolonged multiplexed chemogenetic experiments. Scale bars: 10 μm. (L) Quantification of breaks length in the BSCB of SOD1 G93A /ChAT-cre mice subjected to the late and prolonged multiplexed chemogenetic experiments, expressed as % of contralateral. Data are from N = 3 mice. Scale bars: 20 μm. Data information: in (D, F, G, L), data are presented as means ± SD. **P < 0.01, ***P < 0.001, ****P < 0.0001 (unpaired t test).
For the quantification of the astrocytic end-feet coverage of spinal cord vessels, blood vessels were first identified using collagen-IV staining, and a ROI corresponding to the vessels was manually traced; the surface of the vessel occupied by GFP-positive processes was then quantified, and a ratio between the GFP + area and the total vessel area was considered. The same procedure was used for the quantification of AQP4 around the microvessels. For quantitative analysis, a minimum of 8-10 artifact-free sections per mouse was analyzed, and from each 16-20 vessel, stretches were taken into consideration.

Statistical analysis
One-way ANOVA with Bonferroni correction for multiple comparisons was applied for the comparison of multiple groups. For the comparison of multiple groups with various conditions, two-way ANOVA and nonparametric statistical analysis were applied. The unpaired t test was used to determine if the difference between two groups is significant. Statistical analysis was performed with Prism software (GraphPad6 and 8). All values were expressed as mean ± SD unless otherwise indicated. Statistical significance was set at P < 0.05 before multiple comparison correction.