Neurons diversify astrocytes in the adult brain through sonic hedgehog signaling

See allHide authors and affiliations

Science  19 Feb 2016:
Vol. 351, Issue 6275, pp. 849-854
DOI: 10.1126/science.aab3103

Glial cell properties dictated by neurons

Neurons in the brain coexist with astrocytes, a type of glial cell, which help support many functions of their neighboring nerve cells. Farmer et al. now show that the support goes both ways (see the Perspective by Stevens and Muthukumar). They explored the influence of neurons on two specialized types of astrocytes in the mouse cerebellar cortex. The neurons produced the morphogen known as Sonic Hedgehog. Hedgehog signaling adjusted distinctive gene expression within the two astrocyte cell types. Thus, mature neurons appear to promote and maintain specific properties of associated astrocytes.

Science, this issue p. 849; see also p. 813


Astrocytes are specialized and heterogeneous cells that contribute to central nervous system function and homeostasis. However, the mechanisms that create and maintain differences among astrocytes and allow them to fulfill particular physiological roles remain poorly defined. We reveal that neurons actively determine the features of astrocytes in the healthy adult brain and define a role for neuron-derived sonic hedgehog (Shh) in regulating the molecular and functional profile of astrocytes. Thus, the molecular and physiological program of astrocytes is not hardwired during development but, rather, depends on cues from neurons that drive and sustain their specialized properties.

Astrocytes have fundamental roles in nearly all aspects of brain function, including extracellular ion and neurotransmitter homeostasis, neurometabolism, and cerebrovasculature control (13). Prime examples are pH-sensing brainstem astrocytes that mediate respiratory control (4) and AMPA (α-amino-3-hydroxy-5-methyl-4 isoxazolepropionic acid)–type glutamate receptor–expressing glia that function in cerebellar motor learning (5). Distinct patterns of transcription (6, 7) implicate select genetic programs and possibly distinct signaling mechanisms that establish astrocyte subtypes (2, 8). These processes participate in early developmental patterning events to promote astrocyte heterogeneity in vivo (912).

We explored how molecular features of astrocytes are created and sustained in the mature mouse brain. Because of the complexity of astrocyte heterogeneity in brain areas such as the cerebral cortex, we focused on the cerebellar cortex, which contains two specialized astrocyte types, Bergmann glial cells (BGs) and velate astrocytes (VAs), that have distinct cell positioning, morphology, and molecular composition (13, 14). BGs localized within the Purkinje cell (PC) layer extend processes that enwrap PC dendrites and synapses (Fig. 1A). In contrast, VAs in the granule cell layer (GCL) surround granule cells (GCs) and mossy fiber glomeruli (Fig. 1A) (15). BGs and VAs display distinct, but overlapping, molecular profiles. Although BGs and VAs show comparable expression of genes, including GFAP (glial fibrillary acidic protein), Sox9 [SRY-related high mobility group (HMG)–box gene 9], and GLT1 (glutamate transporter 1), BGs are enriched in AMPA receptors GluA1 and GluA4, and GLAST (glial high-affinity glutamate transporter) (Fig. 1B) (5). VAs, in contrast, have low amounts of GluA1, GluA4, and GLAST and large amounts of the water channel aquaporin 4 (AQP4) (Fig. 1B) (16). Note that components of the sonic hedgehog (Shh) signaling pathway, a developmental morphogen pathway (17, 18), including the Gli1 transcription factor and Shh receptors Patched (patched domain–containing protein) 1 and 2 (Ptch1/2), are also enriched in mature BGs but not VAs (; (Fig. 1B). Ptch2 and Smoothened (Smo) are also expressed by cultured cerebellar astrocytes expressing GLAST (fig. S1).

Fig. 1 Astrocytes in cerebellar cortex, expression of Shh, and Smo deletion.

(A) BG (green) with soma in the Purkinje cell layer (PCL) and processes in the molecular layer (ML), and VAs (magenta) in GCL. (B) Immunofluorescence microscopy for proteins in BGs and VAs. (C) Expression of Shh in PCs expressing calbindin (magenta, asterisks), GCs (right, arrowheads), and interneurons (right, arrow) in Shh reporter mice and after Shh immunolabeling. (D) Removal of Smo through Cre recombination at 5 weeks and analysis 4 weeks later (n = 6 pairs). Scale bars: (B) 30 μm. (C) 15 μm, (D) 10 μm.

In the developing central nervous system (CNS), various cells produce Shh to regulate cell specification, axon guidance, and cell proliferation (17, 18). To identify which cells produce Shh in the mature brain, we used a mouse line that produces tamoxifen-sensitive Cre recombinase from the Shh gene locus (fig. S2A). Cre activation with tamoxifen in >5-week-old mice revealed that PCs, GCs, and interneurons expressed Shh (Fig. 1C and fig. S2, B to E). Immunolabeling for Shh showed localization in neurons, including PCs, and an overall enrichment in the molecular layer (Fig. 1C and fig. S2, F and G). To determine whether Shh signaling regulates mature BGs in vivo, we removed the Shh signal transducer Smo from BGs using controlled activation of Cre with tamoxifen in astrocytes expressing GLAST (GLAST CreERT2) (fig. S3A) (19, 20). Quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) showed a 24% loss of Smo mRNA after Cre activation (fig. S3B), an amount that did not disrupt cerebellar organization or motor performance (fig. S3C). BGs lacking Smo extended processes that enwrapped PC dendrites, detected by staining of calbindin, and spines, seen by staining of metabotropic glutamate receptor 1 (mGluR1) (fig. S4).

Although Smo was not essential for the structure of BGs, Smo regulated expression of molecules that confer BG specialized properties (5). Shh signaling sustained GluA1 expression and prevented expression of AQP4 (Fig. 1D and fig. S3D). Virally expressed Cre in patches of BGs (Fig. 2A) revealed that Smo was needed for expression of GluA1, GluA4, GLAST, the inward rectifying potassium channel Kir4.1 (21), and Ptch2 (Fig. 2, B and C, and fig. S5) (22). This loss was accompanied by an increase in the amount of AQP4 (Fig. 2, B and C). No changes were observed for GLT1 or overall anatomy (figs. S5 and S6). To assess physiological changes to BGs, we used AMPA uncaging to elicit AMPA-receptor responses in BGs (Fig. 2D). This showed reduced AMPA receptor–mediated currents after Smo loss (Fig. 2E). To determine whether Shh expressed by PCs maintains BG gene expression, we removed Shh from PCs using Cre (fig. S7A). BGs next to PCs lacking Shh had decreased amounts of GluA1, Kir4.1, and GLAST and increased amounts of AQP4 (Fig. 2F and fig. S7B), with no disruption to the presence or position of cells containing SOX9 or the expression of GLT1 (fig. S8).

Fig. 2 Purkinje cells diversify BGs in the mature cerebellum.

(A) (Left) Fluorescence microscopy of membrane-targeted green fluorescent protein (GFP) (mGFP, green) in the cerebellum after viral expression of Cre. (Right) BGs that express mGFP or not (mTom; magenta). (B and C) Smo loss in BGs (green; virus delivered at >5 weeks, analyzed 4 weeks later) [Ptch2 (n = 3 pairs), GluA1 (n = 5), GluA4 (n = 3), GLAST (n = 3), Kir4.1 (n = 3), and AQP4 (n = 5)]. Protein expression determined by colocalization with mGFP reference protein. (D) (Left) Setup to elicit AMPA receptor currents in BGs. (Right) Patched BGs showing the location of the compound (N)-1-(2-nitrophenyl)ethylcarboxy-(S)-α-1-(2-nitrophenyl)ethylcarboxyamino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (NPEC-AMPA) puff pipette and laser pulse (asterisk). (E) AMPA receptor currents in BGs. (Left) Representative trace and quantification of uncaging-evoked AMPA receptor currents in cells with Smo (n = 7; Smo+/+) or not (n = 8; Smoc/c). (F) Immunofluorescence detection of GluA1 and AQP4 in BG after loss of Shh from PCs (n = 4 pairs). Error bars represent SEM. Student’s t test. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001. Scale bars: (A) 300 μm, (B), (D), and (F) 40 μm.

Cerebellar VAs are exposed to lower amounts of Shh than BGs, as indicated by Shh immunolabeling (Fig. 1C) and the low amounts of Shh receptor Ptch2, which is positively regulated by Shh signaling (Fig. 1B). To test whether the Shh pathway could control VAs, constitutively active Smo (SmoM2) was expressed in VAs under the control of Cre (fig. S9A) (23). Virally and genetically induced expression of SmoM2 in VAs reduced amounts of AQP4 and increased GluA1, GLAST, and Kir4.1 (Fig. 3, A to C, and fig. S9, B to D) without affecting cell proliferation (fig. S10). To determine whether increasing the Shh pathway allowed VAs to obtain an mRNA profile resembling that of BGs, we performed RNA sequencing (RNA-seq) on small groups of VAs and BGs individually isolated from fresh brain slices (Fig. 3D and fig. S11). This identified 415 mRNAs that significantly distinguish control VAs from BGs. Hierarchical clustering analysis based on these mRNAs showed that SmoM2-expressing VAs (SmoM2 VAs) showed greater similarity to BGs than control VAs. Genes strongly expressed in control VAs like Edn1 (endothelin 1) and Tlr2 (Toll-like receptor 2) are substantially reduced in SmoM2 VAs, whereas genes expressed in BGs like Anxa7 (annexin A7) are up-regulated in SmoM2 VAs (Fig. 3, E and F). Thus, Shh signaling drives specific changes in VAs, which causes them to acquire a molecular profile that is intermediate between a BG and a VA.

Fig. 3 VAs acquire BG-like profiles upon Shh signaling.

(A) Immunofluorescence detection of GluA1 and AQP4 in VAs upon SmoM2 and Tomato expression after viral Cre expression (>5-week-old mice) (n = 4 pairs). (B) Detection of GluA1 in VAs expressing SmoM2 (white arrowheads) or not (gray and magenta arrowheads). (C) Quantification of GluA1, GLAST, and Kir4.1 in VAs through fluorescence colocalization with Tomato reference protein (n = 4). One-way analysis of variance (ANOVA) with Tukey’s. (D) Experimental steps for single-cell RNA-seq. (E) Dendrogram representing a hierarchical clustering of gene expression distances between samples used in the single-cell RNA-seq experiment. Histogram shows a pseudo-color representation of the Euclidean distance matrix (from dark blue for zero distance to white for large distance). (F) Gene expression heat map from 415 differentially expressed genes identified in BGs compared with VAs. Colors reflect relative differences of each gene (y axis) for each sample (x axis). Unsupervised clustering trees are shown, and histogram shows relative expression level (white for lower expression and blue for higher expression). Error bars represent SEM. *P ≤ 0.05, ***P ≤ 0.001. Scale bars: (A) 20 μm, (B) 30 μm.

We also tested if Shh signaling regulates BGs during development by altering Shh signaling at postnatal day 2 (P2) and analyzing mice at P15 (fig. S12). Smo deletion decreased amounts of GluA1 and increased amounts of AQP4 in developing BGs (fig. S13, A and B), whereas SmoM2 expression increased abundance of GluA1, GluA4, and Kir4.1 in VAs (fig. S13, C and D). SmoM2 expression led to a small, significant increase in proliferating glia that contained Ki67, which suggested that Shh signaling promotes cell division during early stages (fig. S14). Thus, cerebellar astrocytes use the Shh pathway at developing and adult stages to establish and sustain their molecular features.

Shh is expressed by neurons in several adult brain regions (2426). To determine whether astrocytes in these areas respond to the Shh pathway, we expressed SmoM2 in mature astrocytes (fig. S15) and examined the expression of proteins, including Kir4.1, which shows heterogeneous expression (Fig. 4 and fig. S15). In the hippocampus, SmoM2 expression increased Kir4.1 protein in CA1 and dentate gyrus astrocytes and overall Kir4.1 mRNA (Fig. 4, A to C, and figs. S15 and S16). Kir4.1 mRNA was also decreased in Shh haploinsufficient mice, which showed reduced Shh and Gli1 mRNA (Fig. 4D). Kir4.1 up-regulation also increased barium-sensitive Kir4.1 currents, as revealed by changes in the rectification index (Fig. 4, E to H, and fig. S17) (21). Similarly, Kir4.1 mRNA levels were reduced in cultured hippocampal astrocytes exposed to Shh pathway inhibitors (fig. S18) (27, 28). GLAST, GluA1, and GluA4 expression were unaffected (figs. S19 to S21). In cortical astrocytes, removal of Smo reduced Kir4.1 protein (fig. S22, A and B), whereas expression of SmoM2 increased astrocyte Kir4.1 protein (fig. S22, C and D, and fig. S23). This was consistent with decreased and increased Kir4.1 mRNA in the cortex of Shh haploinsufficient and SmoM2 mice, respectively (fig. S22, E and F). GLAST and AQP4 mRNAs remained unchanged in SmoM2 and Shh haploinsufficient mice (figs. S22, S24, and S25).

Fig. 4 Shh signaling controls hippocampal astrocytes.

(A) Astrocytes containing SmoM2 (white arrowheads) or not (magenta arrowheads). Cre was activated at 5 weeks with mice analyzed 4 weeks later. (B) Quantification of Kir4.1 expression in CA1 and dentate gyrus through fluorescence colocalization with Tomato reference protein (n = 4 pairs). One-way ANOVA with Tukey’s. (C) mRNA levels in hippocampus (>5 weeks; 2 weeks after Cre induction; n = 7 control and 8 SmoM2 mice). Student’s t test. (D) mRNA levels in Shh haploinsufficient mice (>5 weeks; n = 6 pairs). Student’s t test. (E to H) Ba2+-sensitive Kir4.1 currents in astrocytes 2 weeks after SmoM2 expression. (E) (Left) Voltages were stepped from –160 to 0 mV, with initial step to inactivate non-Kir K+ currents. (Middle, right) Representative currents before (a) and after (b) Ba2+, with Kir4.1 current determined by subtraction (a – b). (F and G) Hierarchical clustering analysis grouped SmoM2(–) (n = 6 cells) with controls (Tom, n = 10) but segregated SmoM2(+) (n = 8). (H) SmoM2(+) Ba2+-sensitive current-voltage curve differed below the reversal potential. Wilcoxon-Mann-Whitney test. Error bars represent SEM. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001. (A) Scale bars, 500 μm and 20 μm.

We found that astrocytes depend on cues from mature neurons to control their complex molecular profile in vivo. This challenges the concept that astrocytes contain hardwired molecular and physiological programs that are fully determined during development and indicates that neurons communicate with astrocytes to actively regulate their local environment in the brain. Neurons use Shh to control the properties of astrocytes and thus extend its role beyond cell proliferation, specification, and axon guidance during CNS development (17, 18). Surprisingly, astrocytes across brain regions use Shh signaling differently. Cerebellar BGs use Shh signaling to promote glutamate detection (GluA1 and 4) and recovery (GLAST), as well as potassium homeostasis (Kir4.1) (21). This is presumably related to the dense glutamatergic inputs onto PCs in the molecular layer. Note that cerebellar VAs acquire features of the BG transcriptome upon Shh signaling. Cortical and hippocampal astrocytes, in contrast, use Shh signaling for more selective regulation of Kir4.1. It is possible that neurons release an array of factors, including Shh, to create astrocyte complexity in the mature brain (6, 7). These factors likely cooperate with developmental patterning events to generate astrocyte heterogeneity (10, 11, 29) and ultimately ensure that astrocytes are properly specialized for the needs of local neural circuits (30).

Supplementary Materials

Materials and Methods

Figs. S1 to S25

Tables S1 to S5

References (3140)

References and Notes

  1. Acknowledgments: Glast-CreERT2 mice are available from M. Götz under a material transfer agreement with Helmholtz Zentrum München–Deutsches Forschungszentrum für Gesundheit und Umwelt. This work was supported by the Canadian Institutes of Health Research (FDN 143337 to C.W.B., MOP 126137/NIA 288936 to P.J.S., and MOP 111152/MOP 123390 to K.K.M.); Natural Sciences and Engineering Research Council of Canada (DG 418546-2 to P.J.S. and 408044-2011 to K.K.M.); Canada Research Chairs Program (F.C., C.E., and K.K.M.); Brain Canada/W. Garfield Weston Foundation (F.C. and K.K.M.); James McGill Chair Program (C.W.B.); and Canadian Foundation for Innovation (LOF 28331 to P.J.S.). W.T.F was supported by a postdoctoral fellowship from the Research Institute of the McGill University Health Centre. J.P. was supported by a Vanier fellowship. We thank M. Götz for GLAST CreERT2 mice; G. Quesseveur for help with tissue processing; L. Li for mouse technical assistance; S. Scales (Genentech) for Shh antibody; T. Alves-Ferreira for ImageJ support; A. Montpetit, A. Staffa, and staff at Genome Quebec and the Research Institute of McGill University Hospital Centre Molecular Imaging Platform for support and use of instrumentation; and E. Ruthazer and D. van Meyel for helpful feedback on the manuscript. The authors declare no conflicts of interest.
View Abstract

Navigate This Article