A Functionally Defined In Vivo Astrocyte Population Identified by c-Fos Activation in a Mouse Model of Multiple Sclerosis Modulated by S1P Signaling: Immediate-Early Astrocytes (ieAstrocytes)

Visual Abstract


Introduction
Neuroinflammation in the central nervous system (CNS), as epitomized by multiple sclerosis (MS), involves activation of resident immune-competent cells, including astrocytes (Cekanaviciute and Buckwalter, 2016), along with infiltration of peripheral immune cells (González and Pacheco, 2014), which have been associated with multiple CNS pathologies such as MS (Noseworthy et al., 2000), Alzheimer's disease (Acosta et al., 2017), and neuropsychiatric disorders (Trepanier et al., 2016). MS is a prototypical neuroinflammatory disease that also includes demyelination and neurodegeneration in the CNS (Noseworthy et al., 2000). Its cause is unclear; however, in animal models such as experimental autoimmune encephalomyelitis (EAE), immune cells, particularly CD4 ϩ T cells, infiltrate the CNS to initiate demyelination and neurodegeneration. Both EAE and human MS can be therapeutically treated by the FDA-approved medicine fingolimod (FTY720; Herr and Chun, 2007;Chun and Hartung, 2010;Lee et al., 2010;Chun and Brinkmann, 2011;Noguchi and Chun, 2011).

Significance Statement
A new, functionally defined in vivo subpopulation of astrocytes termed immediate-early astrocytes (ieAstrocytes) was identified as the first and predominant CNS cell type showing c-Fos activation, in an animal model of multiple sclerosis (MS). ieAstrocytes track with disease severity. The approved MS drug fingolimod (FTY720, an S1P receptor modulator) reduced ieAstrocyte formation as well as clinical disease. Agents that prevent or reduce ieAstrocytes could represent a new strategic target for treating MS and other neuroinflammatory and neurodegenerative brain disorders.
through S1P receptors, engaging four of the five known subtypes (S1P 1 , S1P 3 , S1P 4 , and S1P 5 ; Kihara et al., 2015). In particular, S1P 1 is thought to be mechanistically important in MS in that a functional antagonism of S1P 1 reduces lymphocyte egress from secondary lymphoid organs, resulting in a decreased number of pathogenic lymphocytes entering the CNS (Mandala et al., 2002;Arnon et al., 2011). Complementing these immune effects, a direct CNS mechanism of action (MOA) for fingolimod has been implicated , which also involves functional antagonism of S1P 1 expressed by astrocytes, based on similar results observed following the genetic removal of S1P 1 from astrocytes, which both resulted in reduced EAE severity and elimination of FTY720 efficacy, despite intact lymphocyte trafficking effects . S1P signaling is also implicated in controlling astrocyte functions that include gap junction formation (Rouach et al., 2006), migration (Mullershausen et al., 2007), cytokine production (Dusaban et al., 2017), and nitric oxide production (Colombo et al., 2014) via S1P 1 and/or S1P 3 , all of which support S1P-mediated roles for astrocytes in EAE and MS.
To identify CNS cell types altered during neuroinflammatory processes, we developed an unbiased in vivo screen based on transcription of c-Fos (Bullitt, 1990), an immediate-early gene (IEG) that is rapidly transcribed in response to cellular stimuli, independent of de novo protein synthesis (Lau and Nathans, 1987). c-Fos can be activated in astrocytes (McNaughton and Hunt, 1992;Anderson et al., 1994;Morishita et al., 2011;Yester et al., 2015), microglia (Eun et al., 2004), and oligodendrocytes (Muir and Compston, 1996), as well as within neurons associated with memory (Matsuo et al., 2008). We adapted a conditional c-Fos reporter mouse that historically marks cells with nuclear GFP, which could be followed in four-dimensional (3D over time) analyses. Remarkably, more than 95% of disease-activated cells in vivo were astrocytes that we named immediate-early astrocytes (ieAstrocytes). Here, we describe the in vivo screen used to identify the ieAstrocytes produced by EAE challenge, including their temporal and spatial patterns and their quantitative modification by S1P receptor signaling, demonstrated through the use of genetic and pharmacological interventions.

Histologic analyses (iDISCO)
Immunolabeling-enabled three-dimensional imaging of solvent-cleared organs (iDISCO) was employed to visualize GFP-H2B in intact SCs (Renier et al., 2014). Mice were euthanized with an overdose of isoflurane anesthesia, then fixed by intracardiac perfusion of PBS followed by 4% PFA in PBS. SCs were dissected out and postfixed overnight in 4% PFA in PBS; washed in PBS, 50% methanol in PBS, 80% methanol in PBS, and 100% methanol; and bleached in 20% DMSO in methanol containing 5% H 2 O 2 at 4°C. Samples were washed again in 100% methanol, 20% DMSO in methanol, 80% methanol in PBS, 50% methanol in PBS, PBS, and PBS containing 0.2% Triton X-100 and incubated at 37°C in PBS containing . c, GFP-H2B ϩ population in the spinal cord was determined by FCM (n ϭ 4 animals, mean Ϯ SEM, ‫ء‬p Ͻ 0.05, one-way ANOVA with Bonferroni's post hoc test). d-l, Representative spinal cord coronal sections. In the naive SC, DOX inhibited extra-experimental tagging of GFP ϩ cells. Virtually no GFP-H2B was observed (d-f). On DOX removal, there were sparse and dispersed GFP ϩ cells in the naive SC over 5 d (g-i). EAE induced robust GFP signals in the SC at 5 days post-onset (dpo) that were clustered in lesions and along the periphery (j-l). Scale bar, 500 m. Figure 2. c-Fos activation is temporally and spatially propagated in EAE SC. a-h, Representative 3D imaging of lower (lumbar region) SCs in naive (a and b) and EAE-induced reporter mice of 1 (c and d), 3 (e and f), and 5 (g and h) days post-onset (dpo, n ϭ 2-5 animals for each dpo). a, c, e, g, 3D-reconstructed images. b, d, f, h, Coronal sections. Scale bar, 500 m. i, The number of GFP-H2B ϩ nuclei in the SCs (mean Ϯ SEM, ‫ء‬p Ͻ 0.001 versus naive, Kruskal-Wallis test with Dunn's post hoc test, data were compiled from 273, 149, 185, and 381 z-slices from 3, 2, 2, and 4 animals, respectively). j, Distribution of the number of GFP-H2B ϩ nuclei per z-slice as shown in i. Lines represent Gaussian fits to the data. k, Average size of GFP-H2B ϩ signals (mean Ϯ SEM, ‫ء‬p Ͻ 0.001 versus naive, Kruskal-Wallis test with Dunn's post hoc test, data were compiled from 273, 149, 185, and 381 z-slices from 3, 2, 2, and 4 animals, respectively). l, Distribution of the average size of GFP-H2B ϩ signals per z-slice shown in k. Lines represent Gaussian fits to the data. Theory/New Concepts 0.2% Triton X-100/20%DMSO/0.3 M glycine. Samples were blocked in PBS containing 0.2% Triton X-100/10% DMSO/6% goat serum at 37°C, washed in PBS containing 0.2% Tween 20 and 10 g/ml heparin (PTwH), and incubated in a rabbit anti-GFP antibody (1:1000 dilution, MBL International Cat# 598, RRID:AB_591819) in PTwH at 37°C. Samples were washed in PTwH, then incubated in anti-rabbit Alexa Fluor 488 (1:1000 dilution, Thermo Fisher Scientific Cat# A-11008, RRID: AB_143165). Samples were washed again in PTwH before tissue clearing. SC samples were incubated in 50% v/v tetrahydrofuran/H 2 O (THF, Sigma), 70% THF in H 2 O, 80% THF in H 2 O, 100% THF, dichloromethane (Sigma), and finally dibenzyl ether (DBE). Cleared SCs were placed in a microscope chamber made with two to three stacked Fastwells (Research Products International), filled with DBE, and coverslipped. The intact SCs were imaged on a Nikon A1 ϩ confocal microscope using z-steps of 10 m under 10ϫ objective. NIS-Elements (v4.3, RRID:SCR_014329) was used to three-dimensionally reconstruct the entire SC. Obtained images were analyzed with ImageJ (NIH, RRID:SCR_003070) to count the number of signals, areas, and average area of signals.

Statistical analysis
Results are expressed as means Ϯ SEM. Data were analyzed statistically by means of ANOVA with indicated post hoc tests as appropriate, using GraphPad PRISM software (RRID:SCR_002798). Values of p Ͻ 0.05 were considered to be statistically significant.

A new, in vivo c-Fos screen detects CNS cells activated by neuroinflammation
We hypothesized that EAE would induce c-Fos expression in key, disease-relevant cells. An unbiased screen was developed using the c-Fos-inducible and doxycycline-regulated TetTag cFos system that labels cell nuclei with GFP permanently (Fig. 1a). This system was used previously to identify neurons involved in learning and memory (Matsuo et al., 2008). TetTag cFos reporter mice, immunized with MOG 35-55 peptide, showed robust GFP signals in the SC 5 d after removal of doxycycline (Fig. 1). The increase was detected only in symptomatic EAE mice (Fig. 1c), indicating that EAE insults produced c-Fos activation in the CNS. To understand the spatio-temporal c-Fos activation pattern, EAE SCs were processed by a tissue-clearing technique, iDISCO (Renier et al., 2014), . Control SC from naive mice displayed sparse GFP signals even 5 d after doxycycline removal (Video 1). After 1 d post-onset (1 dpo), GFP ϩ nuclei appeared in the periphery of white matter tracts near associated blood vessels (Video 2). The GFP signals expanded along the rostral-caudal axis and penetrated into the white matter parenchyma over time, appearing most prominently near anatomically known major blood vessels including the anterior, posterior, and posterolateral spinal vessels (Fig.  2, Videos 3 and 4). Importantly, the number of GFP ϩ nuclei (Fig. 2i, j) and the average size of GFP ϩ signals (Fig.  2k, l) increased significantly over time, revealing that GFP ϩ nuclei aggregated to form clusters in EAE SCs at 5 dpo. These results identified robust c-Fos activation in cells near major CNS blood vessels in peripheral white matter tracts, followed by inward expansion centrally, into gray matter.

c-Fos ؉ cells are astrocytes: ieAstrocytes
To determine the identities of GFP ϩ cell types, immunolabeling for various cellular markers was performed on 5-dpo SC sections. Strikingly, a vast majority of GFP ϩ nuclei (95.09 Ϯ 0.73%, n ϭ 3) were found to colabel with glial fibrillary acidic protein (GFAP; Fig. 3). By contrast, Ͻ1% of GFP ϩ nuclei (0.66% Ϯ 0.14%, n ϭ 3) colocalized with a neuronal nuclear marker (NeuN). Other assessed cell types included microglia, oligodendrocytes, and T lymphocytes, none of which showed colocalization with GFP signals (Fig. 3), whereas GFP ϩ CD3 ϩ T cells were found outside of the CNS within the peripheral lymph nodes of EAE mice, demonstrating their ability to report c-Fos (data not shown). The dominance of EAE-induced GFP ϩ astrocytes that reported transcription of the c-Fos IEG during disease development supported a functional classification for these astrocytes, which were therefore called immediate-early astrocytes (ieAstrocytes). At any given time following EAE challenge, ieAstrocytes could be distinguished from cells that did not show evidence of c-Fos activation, quiescent astrocytes (qAstrocytes).
Flow cytometric (FCM) analysis of ieAstrocytes (DAPI-gated, NeuN -GFP ϩ populations) from the SCs of EAE mice revealed a significant decrease of ieAstrocytes (Fig. 4j) and a linear correlation of increasing ieAstrocytes with worsening clinical scores (Fig. 4k). These results link ieAstrocyte formation to EAE disease progression, both of which can be modified by S1P 1 signaling.

Discussion
The present study identified a functionally defined in vivo population of astrocytes, ieAstrocytes, as the earliest and predominant cell type responding to EAE insult. Surprisingly, other cell types were not functionally activated to a similar extent within the CNS. Moreover, ieAstrocyte formation increased with disease severity, and in part required S1P-S1P 1 signaling.
IEGs as typified by c-Fos have been used as functional markers of cellular activation (Herrera and Robertson, 1996), particularly in the nervous system, where c-Fos was used as a transynaptic marker for neuronal activity (Bullitt, 1990), and most commonly relied on immunolabeling. In the present study, we used a TetTag c-Fos reporter system that is remarkable in tracking historical neural activity by labeling cells with stable GFP, compared with merely capturing momentary snapshots by classical c-Fos immunolabeling. This system historically traced the cells that experienced c-Fos expression during challenges, which identified a functional subset of neurons in learning and memory (Matsuo et al., 2008;Tayler et al., 2011) and functional astrocytes in EAE (Fig. 2).
Although several studies identified c-Fos immunoreactivity in neurons with exclusion of c-Fos expression in astrocytes (Staiger et al., 2002;Wang et al., 2009;Lecrux et al., 2011), it was demonstrated in cultured astrocytes stimulated with excitatory amino acids (McNaughton and Hunt, 1992) and S1P (Yester et al., 2015), as well as in in vivo astrocytes, such as in electrical sensorimotor cortex stimulation (Morishita et al., 2011) and Alzheimer's brains (Anderson et al., 1994). Moreover, results from direct activation of engineered G␣i-coupled GPCRs expressed Video 5. 3D video of WT fos EAE spinal cord at 5 dpo shown in Fig. 4a, b. PRINT [ View online] Video 6. 3D video of S1P 1 -AsCKO fos EAE spinal cord at 5 dpo shown in Fig. 4c, d. PRINT [ View online] in in vivo astrocytes using designer receptors exclusively activated by designer drugs (DREADDs; Chai et al., 2017), concur with our results that G␣i-coupled S1P 1 signaling functionally activates c-Fos in astrocytes in vivo (Fig. 3).
Astrocytes are the most abundant cells in the CNS (Burda and Sofroniew, 2014;Farmer and Murai, 2017). They have been widely considered as bystanders in neuroinflammatory diseases (Rossi et al., 2007), responding secondarily under disease conditions to become reactive astrocytes that were originally defined by their hypertrophic morphology and increased GFAP expression, compared with quiescent qAstrocytes (Burda and Sofroniew, 2014;Hubbard and Binder, 2016). A recently proposed classification of A1 reactive astrocytes added transcriptomic signatures to the reactive astrocyte populations (Liddelow et al., 2017). By contrast, ieAstrocytes are defined (1) functionally (via c-Fos expression) and (2) tem-porally (in vivo during EAE development and progression) (Fig. 4). Considering the nature of immediate-early and transient c-Fos expression (Herrera and Robertson, 1996), ieAstrocytes appear to differ from reactive astrocytes, possibly representing an intermediate/transition state between qAstrocytes and reactive astrocytes, which moreover is partly driven by S1P signaling (Fig. 4). A possible scenario is that during EAE development, bloodborne S1P carried by albumin and apolipoprotein M (Blaho et al., 2015) at sufficient concentrations (ϳM) to activate S1P receptors (Tsai and Han, 2016) extravasates across a disrupted blood-brain barrier (Alvarez et al., 2011) to reach perivascular astrocytes in the spinal cord ( Fig. 2 and Videos 1-7), resulting in direct activation of qAstrocytes to initiate ieAstrocyte formation. An alternative, but not mutually exclusive scenario, is the formation of A1 reactive astrocytes requiring microglial cytokine production (IL-1␣, TNF␣, and C1q) to activate astrocytes secondarily (Liddelow et al., 2017), to give rise to both ieAstrocytes and reactive astrocytes. The determination of the involved activation pathways will require further study.
Therapeutic reduction of S1P 1 signaling is thought to reduce pathogenic lymphocyte trafficking into the brain, which has been proposed as the immunologic MOA of fingolimod (Mandala et al., 2002;Chun and Hartung, 2010;Arnon et al., 2011;Cohen and Chun, 2011;Chun et al., 2018). However, a direct CNS MOA involving astrocytes Groves et al., 2013) receives further support through the dominant presence of ieAstrocytes, whose increasing numbers correlated with disease severity (Fig. 4k), and from the fact that genetic removal or pharmacological inhibition by fingolimod of S1P 1 signaling Groves et al., 2013) reduced both ieAstrocytes and disease severity. Notably, complete S1P 1 removal did not eliminate ieAstrocytes, indicating that other factors must be involved in their activation. In this light, other MS drugs that have been proposed to access a direct CNS MOA distinct from S1P receptor modulation in EAE, including interferon ␤ (Rothhammer et al., 2016) and dimethyl fumarate (Al-Jaderi and Maghazachi, 2016), might also reduce ieAstrocyte formation. Although these relationships remain to be determined, they theoretically support combination therapies that could more completely prevent ieAstrocyte formation. A next-generation S1P receptor modulator, siponimod (BAF312), that also engages S1P 1 (and S1P 5 ), showed positive results in a phase III secondary progressive MS trial (Kappos et al., 2018), which is consistent with a CNS MOA involving ieAstrocytes, while other related agents like ozanimod (Cohen et al., 2016), etrasimod (Buzard et al., 2014), and ponesimod (Bolli et al., 2010;Vaclavkova et al., 2014)-having similar, albeit distinct, S1P receptor engagement properties-may also have direct CNS effects in MS through ieAstrocytes. Overall, our results support a previously unrecognized pathogenic role of functionally defined astrocytes in vivo, which could reveal new disease mechanisms and therapeutic targets accessed through ieAstrocytes.