ATP mediates neuropathic pain in neuromyelitis optica via microglial activation

establish We established an NMOSD pain model by injecting anti-AQP4 recombinant autoantibodies (AQP4-Ab) from NMOSD patient plasmablasts into rat spinal cords. We performed transcriptome analysis and pharmacological inhibition to elucidate the core mechanism of allodynia in the model. AQP4-Ab: anti-AQP4 recombinant autoantibodies; DAMPs: damage-associated molecular patterns; DEG: Differentially expressed gene; HEK293: Human Embryonic Kidney cells 293; HEK-AQP4: HEK293 cells expressing AQP4; MS: multiple sclerosis; NMOSD: neuromyelitis optica spectrum disorder; OND: other neurological disease; PCA: principal component analysis; rAQP4 IgG: recombinant monoclonal AQP4 antibody


Abstract Background
Intractable neuropathic pain is a common symptom of neuromyelitis optica spectrum disorder (NMOSD). However, the underlying mechanism of NMOSD pain remains to be elucidated. The aim of this study was to establish a novel animal model of NMOSD pain and to investigate its pathogenic mechanism.

Methods
We established an NMOSD pain model by injecting anti-AQP4 recombinant autoantibodies (AQP4-Ab) from NMOSD patient plasmablasts into rat spinal cords. We performed transcriptome analysis and pharmacological inhibition to elucidate the core mechanism of allodynia in the model.

Results
Development of mechanical allodynia was con rmed in the NMOSD pain model. AQP4-Ab mediated extracellular ATP release in vitro, and pharmacological inhibition of ATP receptor reversed mechanical allodynia in the NMOSD pain model. Furthermore, transcriptome analysis revealed microglial activation and elevated levels of IL-1β in NMOSD spinal cord. Inhibition of microglial activation and neutralization of IL-1β also attenuated neuropathic pain in the NMOSD rat model. In human patients, CSF ATP concentration was signi cantly higher in the acute and remission phase of NMOSD than in multiple sclerosis or other neurological disorders.

Conclusion
A novel NMOSD pain model was established. ATP, microglial activation, and IL-1β secretion orchestrate the pathogenesis of NMOSD neuropathic pain.

Background
Neuromyelitis optica spectrum disorder (NMOSD) is an autoimmune in ammatory disease of the central nervous system that presents with optic neuritis or longitudinally extensive myelitis and is associated with various neurological symptoms [1]. Approximately 80% of patients with NMOSD experience intractable pain, and their quality of life is severely affected [2]. Current therapeutics for NMOSD pain, including antiepileptic agents, anti-spasticity medications, and opioids, provide insu cient relief of the symptoms [3]. However, no models for investigating the mechanisms of NMOSD pain have yet been reported [2]. Consequently, the detailed pathogenesis of NMOSD pain remains to be elucidated.
Anti-aquaporin4 autoantibody (AQP4-Ab) plays a pathogenic role in NMOSD [4][5][6]. We showed previously that mitochondrial DNA (mtDNA), a damage-associated molecular pattern (DAMP), is released by dying astrocytes in an AQP4-Ab-dependent manner and further augments in ammation via innate immune receptors [7]. Several DAMPs, including ATP, are related to neuropathic pain [8], and the pivotal roles of ATP and purinergic receptors have been demonstrated in a peripheral neuropathic pain model [9].
In this study, we sought to establish an NMOSD pain model using patient-derived monoclonal antibodies [10], and to further clarify the role of the purinergic pathway in NMOSD pain.

Methods
Patient information and sample collection CSF was obtained from patients with AQP4-Ab-positive NMOSD in the acute (n = 17) or remission phase (n = 18), MS in the acute (n = 15) or remission phase (n = 6), or other neurological diseases (ONDs, n = 56). All NMOSD subjects were diagnosed according to the 2015 NMOSD diagnostic criteria, and all MS patients ful lled the 2010 McDonald criteria. ONDs included Guillain-Barré syndrome (GBS) (n = 6), amyotrophic lateral sclerosis (ALS) (n = 27), Parkinson disease (PD) (n = 6), idiopathic normal pressure hydrocephalus (iNPH) (n = 5), cervical spondylosis (n = 5), and somatic symptom disorders (n = 7). The acute phase of NMOSD was de ned as a sudden appearance of new neurological symptoms, and CSF during the acute phase was collected before or within 24 h from the start of treatment (high-dose intravenous methylprednisolone or plasmapheresis). CSF in the remission phase was collected after symptoms resolved following treatment for the acute phase. Informed consent was obtained from each patient. This study was approved by the ethics committee of Osaka University Hospital (permit number 12091-6).

Surgical procedures
Female Lewis rats (age: 8 weeks; body weight: 200-250 g) purchased from the Charles River Laboratories Japan (Yokohama, Japan) were anesthetized with a mixture of Dormicam (4 mg/kg), Vetorphale (5 mg/kg), and Domitor (4 mg/kg), and laminectomy was performed at the thoracic 10 (Th10) vertebral level, where the spinal cord was exposed. A microsyringe (Hamilton, 75RN NEURO SYRINGE, Cat. No.4015-63014) was inserted 1.5 mm at Th10 and used to infuse patient-derived AQP4 recombinant antibody [10] or control IgG [Nordic-MUbio, Human IgG1 lambda G1m(f)] (20μg), 1 µL of normal human serum, and indocyanine green (Tokyo Chemical Industry, I0535). Total volume was 4 μL. Rats in the TNP-ATP-or anti-IL-1β antibody-treated group received a mixture of 1 μg TNP-ATP (Sigma-Aldrich, SM0740) or 1 μL anti-IL-1β antibody (BioXCell, BE0246). Rats in the minocycline-treated group received three doses of 80 mg/kg minocycline (Sigma-Aldrich, USA, M9511) on postoperative days 0, 1, and 2. All experimental procedures were performed according to the protocols approved by the Animal Care and Use Committee of Osaka University School of Medicine and the US Public Health Service Policy on Humane Care and Use of Laboratory Animals.

Behavioral studies
To assess mechanical sensitivity, calibrated von Frey laments (1.0-26 g; Aesthesio, DanMic Global) were applied to the plantar surfaces of the hindpaws of NMOSD and control rats. The rats were placed in a plastic cage with a wire mesh bottom that allowed access to their paws. Behavioral acclimatization was allowed for 1 h until cage exploration and grooming activities ceased. The area tested was the midplantar hindpaw. Each lament was tested ve times per paw, and the threshold was de ned as three or more withdrawals observed in ve trials. Heat hyperalgesia was assessed using a Hargreaves radiant heat apparatus (Ugo Basile). The heat source, a mobile infrared photobeam, was positioned under the plantar surface of the hind paw. The intensity of the heat stimulus was adjusted to 50, and the cutoff was set to 20 s. Cell cultures HEK293 cells transfected with or without M23-human AQP4 expression plasmid (GeneCopoeia, Rockville, MD, USA) were cultured in Dulbecco's Modi ed Eagle's Medium containing 10% fetal bovine serum and 1% penicillin-streptomycin.
Quanti cation of ATP HEK293 cells were seeded in 24-well culture plates and stimulated with DMEM with or without AQP4 recombinant antibody [10] containing 0, 5, or 10% normal human serum at 37°C. Culture supernatants were collected 2 h after antibody stimulation and centrifuged at 2000 g at 4°C for 10 min. Supernatants were analyzed using the ATP Bioluminescence Assay Kit HS II (Roche, 11699709001). The standard samples and control groups were added to the corresponding wells. CSF samples were centrifuged at 2000 g at 4°C for 10 min. ATP levels in CSF samples of patients were measured using the ATPlite luminescence assay system (PerkinElmer, 6016943). In this assay, the emitted light produced by the reaction of ATP with added luciferase and D-luciferin is proportional to the concentration of ATP. In a 96well white plate, 50 μL of the reconstituted reagent was added to each well, which contained 100 μL of CSF supernatant centrifuged at 2000 g at 4°C for 10 min and equilibrated at room temperature. Luminescence was measured using a luminescent microplate reader (Berthold Technologies, Centro XS3 LB960).

Transcriptome analysis
For transcriptome analysis of spinal cord tissue, NMOSD or control rats were transcardially perfused with ice-cold PBS under anesthesia with i.p. injection of a mixture of Dormicam (4 mg/kg), Vetorphale (5 mg/kg), and Domitor (4 mg/kg). Spinal cords (n = 3 each group) were cut out at the Th10 level, and total RNA was prepared using ISOGEN II (Nippongene). RNA libraries were prepared for sequencing using the TruSeq Stranded mRNA Library Prep kit (Illumina, San Diego, CA, USA). Sequencing was performed on the Illumina HiSeq 2500 platform in a 75bp single-end mode. Sequenced reads were mapped to the rat reference genome sequence (rn6) using TopHat ver. 2.1.1 in combination with Bowtie2 ver. 2.2.3. The raw counts were calculated using Cu inks ver. 2.2.1. Transcriptome analysis was performed using either iDEP.91 [11], the STRING database [12], or Enrichr [13]. For principal component analysis (PCA), normalization and scaling of raw read counts were processed by the EdgeR package of iDEP.91 ( ltered with 0.5 counts per million); among the six samples, 12,453 genes passed the lter. Differentially expressed genes (DEGs) were assessed by DESeq2 (false discovery rate cutoff <0.1, fold change >2) and visually depicted using the STRING database. Heat map of purinergic receptors was calculated with iDEP.91 by normalizing samples against the standard deviation. Using Enrichr-based WikiPathways analysis, Gene Set Enrichment Analysis was performed with upregulated DEGs of NMOSD rats that met the criterion |log 2 (fold-change)| >1. Raw data from this analysis were submitted to Gene Expression Omnibus (GEO) under accession number GSE150598.

Statistics
Data are shown as means ± SEM. Data assessing allodynia and HEK-AQP4 ATP assay were analyzed by one-way or two-way mixed regression analysis for ANOVA model with random-effects of one withinsubjects (the same rat's paired right and left data), with post hoc Bonferroni correction. CSF ATP levels of different groups were analyzed by Welch's T test. Statistical signi cance was de ned as P < 0.05. Asterisks in gures denote P values as follows: *P < 0.05, **P < 0.01.

Results
Administration of rAQP4 IgG to rats recapitulates neuropathic pain of NMOSD To determine whether patient-derived AQP4 antibodies reproduce neuropathic pain observed in NMOSD patients, we rst generated recombinant monoclonal AQP4 antibody (rAQP4 IgG) from NMOSD plasmablasts. Patient-derived rAQP4 IgG was stereotaxically injected into rat spinal cords at the T10 vertebra ( Fig 1A). Loss of GFAP and AQP4 staining were observed in in spinal cord injury site (Fig 1B). Three days after the injection, the rats were tested for mechanical and thermal paw withdrawal responses. Relative to the control group, rats receiving rAQP4 IgG exhibited a lower threshold for mechanical pain in both hind limbs (Fig 1C). By contrast, thermal paw responses did not differ signi cantly between the two groups ( Fig 1D). No apparent motor de cit was observed in both groups. These results demonstrate that patient-derived rAQP4 IgG is capable of recapitulating neuropathic pain of NMOSD in a rat model, as most prominently represented by mechanical allodynia.
rAQP4 IgG promotes extracellular ATP release Among the molecules reported to play pivotal roles in peripheral neuropathic pain, ATP is an essential factor that also functions as a DAMP in immune-mediated cellular damage. Hence, to determine whether rAQP4 IgG promotes ATP release from cells expressing AQP4, we incubated HEK293 cells expressing AQP4 (HEK-AQP4) with patient-derived rAQP4 IgG. After 1 h of incubation, the rAQP4 IgG-treated cells exhibited a balloon-like morphology (Fig 2A to D). Concomitant with the morphological alteration, the level of ATP in the supernatant was elevated in the rAQP4 IgG-treated cells. Importantly, the amount of ATP released was higher in the presence of 10% normal human serum addition than in the presence of 5% serum (Fig 2E). These results indicate that in the presence of complement, rAQP4 IgG induces a balloon-like morphological change in HEK-AQP4 cells, followed by extracellular release of ATP, a key molecule involved in neuropathic pain. Similarly, rAQP4 IgG also induces extracellular ATP release from rat primary astrocytes (Fig. S1).
Inhibition of ATP receptors prevents neuropathic pain of NMOSD model.
Given that our in vitro study showed that rAQP4 IgG induced the extracellular release of ATP, we next assessed whether ATP receptor antagonist would prevent the development of neuropathic pain in the NMOSD rat model. To this end, we locally co-administered TNP-ATP (an antagonist of P2RX1, P2RX3, P2RX2/3, and P2RX4) when rats were injected in the spinal cord with either rAQP4 IgG or control IgG. Three days after injection, the rats were tested for mechanical paw withdrawal responses. In the rAQP4 IgG-treated group, TNP-ATP prevented neuropathic pain symptoms, bringing the threshold of mechanical allodynia to a level comparable to that in the control IgG group (Fig 2F). These results con rmed the complete prevention of pain by TNP-ATP in the NMOSD rat model, and further support the hypothesis that ATP plays a crucial role in the pathogenesis of neuropathic pain in this model.

Transcriptome pro ling of the NMOSD model reveals elevation of IL1B gene expression and activation of innate immune pathways
To further characterize the molecular signature of neuropathic pain in the NMOSD model, we performed RNA sequence on spinal cords of rAQP4-IgG-or control IgG-treated rats. Principal component analysis (PCA) (Fig 3A) revealed a clear difference in gene expression between the two groups. Among the differentially expressed genes, several ATP receptors including P2RX4 were elevated in the NMOSD model ( Fig 3B). Gene Ontology (GO) enrichment analysis revealed that the genes involved in Toll-like receptor signaling, chemokine signaling, IL-5 signaling, and complement cascades, all of which play important roles in human NMOSD pathology, were highly enriched in NMOSD rats (Fig 3C). Furthermore, a network analysis of known and putative protein-protein interactions in the NMOSD model revealed that the upregulated genes are connected, and that some of them are hub genes, e.g., IL1B, which is a core downstream molecule in pain hypersensitivity induced by peripheral nerve injury (Fig 3D). The fold change of IL1B was one of the highest in all genes (Table. S1).

Microglial activation and IL-1β contributes to pain induction in NMOSD model
The transcriptome analysis highlighted the substantial activation of innate immune pathways, especially the elevation of IL1B expression in the NMOSD model. Furthermore, among the components of the molecular network related to pain induction in innate immune responses, IL-1β was previously reported to function as a core factor in this context. Hence, to further clarify whether enhancement of IL-1β contributes to pain induction in the NMOSD rat model, neutralizing antibodies for IL-1β were locally administered into the spinal cord along with either rAQP4 IgG or control IgG. Mechanical paw withdrawal responses were signi cantly recovered in NMOSD rats receiving neutralizing antibodies against IL-1β ( Fig   4A).
Activated microglia play essential roles in neuropathic pain development and are a major source of IL-1β production [7]. Therefore, we next investigated whether inhibition of microglial activation would alleviate neuropathic pain in NMOSD rat model. To this end, minocycline was systemically administered intraperitoneally for 3 days, and then mechanical paw withdrawal responses were evaluated. Immunohistochemical analysis con rmed a reduction in microglial activation in minocycline-treated rat spinal cords (Fig 4B to E). The threshold for mechanical allodynia was comparable among the minocycline-treated, rAQP4 IgG-treated, and control IgG-treated groups (Fig 4F).
Taken together, these results strongly support the hypothesis that microglial activation contributes to pain induction in the NMOSD rat model, and that IL-1β is an essential factor in innate immune-mediated responses.

Elevation of CSF ATP level is a unique feature of patients with NMOSD
To determine whether concentration is speci cally elevated in cerebrospinal uid (CSF) of NMOSD patients, we assessed the ATP level in the CSF of patients with NMOSD, multiple sclerosis (MS), or other neurological diseases (ONDs). CSF ATP levels were signi cantly higher in patients with NMOSD than in those with MS, in both the acute and remission phases. Similarly, CSF ATP levels were signi cantly higher in NMOSD patients than in OND patients. These results suggest that elevation of CSF ATP levels is a unique feature of patients with NMOSD (Fig 5).

Discussion
In this study, we showed that AQP4-Ab mediates extracellular ATP release from astrocytes, and that pharmacological inhibition of P2XR reverses mechanical allodynia in the NMOSD pain model. Furthermore, we found that the CSF ATP concentration was signi cantly higher in the acute and remission phase of NMOSD than in MS and other neurological disorders. Transcriptome analysis revealed microglial activation and elevated IL1B expression in NMOSD spinal cord. Inhibition of microglia activation or neutralization of IL-1β attenuated neuropathic pain in NMOSD rat model. Together, these ndings indicate that ATP, microglial activation, and IL-1β secretion play key roles in NMOSD neuropathic pain.
Rodent models of peripheral neuropathic pain, like spared nerve injury (SNI), are well established in the pain research eld [14]. However, few studies have developed pain models for in ammatory diseases of the CNS. Moreover, no previous model recapitulated the pain symptoms of NMOSD. We previously reported an NMOSD rat model generated by intraperitoneal injection of IgG derived from NMOSD patients [6]. Although these rats exhibited pathological changes characteristic to NMOSD patients, due to their severe clinical symptoms they were not suitable for accurately assessing the pain symptoms of NMOSD. Therefore, in this study we chose to perform intra-spinal injection of patient-derived monoclonal AQP4-Ab [10] so that lesion formation was restricted and the rats were spared from lower limb immobilization. Our model presented mechanical pain but pain related to NMOSD patients is not limited to mechanical allodynia. Spontaneous painful sensations are also common in NMOSD. Thus, it would be more informative to establish a model which represents the full spectrum of pain symptoms in future studies. In our rat model injection of AQP4-Ab was performed at T10 vertebral level of spinal cord (L1-2 spinal cord level), which is a little distant from L4-5 spinal cord where the nociceptive signals from hind paws enter. This apparently discrepant phenomena can be explained by the previous observation that thoracic tissue damage can propagate to activate L4-5 spine microglia [15]. Transcriptome analysis of spinal cords from the new NMOSD model revealed elevation of gene sets associated with NMOSD pathology, including chemokine signaling, IL-5 signaling, and complement cascades. AQP4-Abs attract granulocytes to lesion sites via chemokine signaling, and IL-5 promotes further tissue injury by activating eosinophils [16]. AQP4-Abs also induce complement-dependent cellular cytotoxicity [17]. These results con rmed that the NMOSD rat model established in this study accurately re ects the characteristics of human NMOSD. Notably, our model was established in the absence of AQP4-speci c T cells [18]. Given that administration of AQP4-IgG and complement alone was su cient to reproduce the neuropathic pain of NMOSD, the downstream molecular events ensuing the activation of complement cascade by AQP4-IgG are likely to be involved in the core mechanism of neuropathic pain development in NMOSD.
CSF levels of HMGB1, a DAMP, are higher in NMOSD patients than in MS and ONDs patients [19]. In addition, the mtDNA level is speci cally elevated in NMOSD CSF [7]. Cellular stress and necrosis are thought to be major contributors to DAMPs release. Thus, the higher CSF DAMPs levels in NMOSD presumably re ect the destructive nature of NMOSD pathology, which is mediated by AQP4-Abdependent astrocyte damage. We found that AQP4-Ab mediates extracellular ATP release from HEK293 cells expressing AQP4. ATP is an essential energy source for cells, but recent studies have revealed that it also acts as a DAMP and further in uences cellular migration and pain induction. In peripheral neuropathy models, injured neurons release ATP in the dorsal horn of the spinal cord [20], which is detected by purinergic receptors expressed by microglia, thereby inducing pain. In another model, C. albicans-derived β-glucan promotes ATP release from keratinocytes and also elicits pain symptoms [21]. Therefore, it is reasonable to speculate that ATP released from dying astrocytes following complement-dependent necrosis functions as a chemical mediator to modulate the surrounding CNS environment to elicit neuropathic pain.
In this study, TNP-ATP, a P2RX4 antagonist, attenuated mechanical allodynia in the NMOSD pain model. In addition, transcriptome analysis revealed that several ATP receptors, including P2RX4, were elevated in our NMOSD rat model, highlighting the central role of ATP signaling. Furthermore, rats treated with minocycline exhibited reversal of mechanical allodynia in the rAQP4 IgG-treated group, suggesting that microglia are the pivotal cellular component in the development of neuropathic pain in NMOSD. Multiple reports have highlighted the importance of purinergic signaling and microglial activation in the pathogenesis of peripheral neuropathic pain. Nerve injury induces upregulation of transcriptional factors of IRF8 and IRF5, along with the purinergic receptor P2RX4, exclusively in spinal cord microglia [22,23]. Furthermore, nerve injury-induced allodynia is abolished in P2rx4-knockout mice, and intrathecal injection of P2RX4-stimulated microglia can cause allodynia [24]. In addition, nerve injury induces upregulation of P2rx7 in spinal cord microglia [25]. Interestingly, P2RX7 and P2RX4 form functional interactions through their physical association [26,27]. Activation of P2RX4 and P2RX7 leads to microglial activation via p38 phosphorylation; increases the synthesis and release of various chemical mediators such as TNF-α, IL-1β, IL-6, BDNF, and PGE2; and facilitates the development of neuropathic pain [28].
Notably in regard to the role of IL-1β in NMOSD pain development, P2RX4 regulates P2RX7-dependent IL-1β release [29]. Additionally, microglia are suggested to be the major source of IL-1β in NMOSD pathology [30]. In line with these observations, in this study we revealed that IL-1β constitutes one of the hubs of molecular events occurring in NMOSD rat models by transcriptome analysis, and further showed that IL-1β neutralization alleviates the mechanical allodynia associated with NMOSD. IL-1β acts on neurons to increase glutamate release [31]; in addition, Il1r1-knockout mice do not develop acute pain in a paclitaxelinduced model [32].
One limitation of our study is that our rat model was suitable for pain assessment only at the acute phase of the disease due to the restricted amount of rAQP4 IgG available for injection into the spinal cord. Nevertheless, ATP levels in the CSF of NMOSD patients were elevated not only in the acute phase, but also in the remission phase. Thus, we can speculate that purinergic signaling plays pivotal roles in both the acute and chronic phases of NMOSD pathogenesis, and that it contributes to the establishment of the neuropathic pain associated with the disease.

Conclusion
We clari ed the essential role of ATP signaling, as well as its downstream molecular events, in the pathogenesis of NMOSD neuropathic pain. Our observations suggest that novel therapeutic agents targeting purinergic and IL-1β signaling could be used to treat patients living with the intractable pain of NMOSD.     Minocycline was administered to the rats depicted in (D) and (E). (F) Thresholds of mechanical allodynia assessed in rats receiving either control IgG or rAQP4 IgG with or without minocycline (n = 6 each group). Data are expressed as means ± SEM and analyzed by two-way repeated measures ANOVA with one within-subjects factor [the same rat's paired right (Rt) and left (Lt) data]. **P < 0.01. Scale bar = 100 µm.