Alexander disease (AxD) is a usually fatal neurodegenerative disorder resulting from astrocyte dysfunction (for reviews, see (Brenner et al. 2009; Flint and Brenner 2011; Messing et al. 2012a)) . Its defining feature is the abundant presence in astrocytes of protein aggregates , called Rosenthal fibers, which are especially prevalent in subpial, periventricular and perivascular locations . Historically, AxD has been classified as a pediatric leukodystrophy because it was first and then most often diagnosed in young children, and because in this age group it is accompanied by massive myelination defects. Common clinical manifestations of childhood AxD include failure to meet mental and physical developmental milestones, megalencephaly, seizures , hyperreflexia and pseudobulbar signs, such as difficulty speaking and swallowing. Magnetic resonance imaging (MRI) criteria are highly diagnostic for childhood AxD, with a major feature being striking myelination defects bilaterally in the frontal lobes (van der Knaap et al. 2001) . The classification of AxD was extended to adult cases based on similar findings of profuse astrocytic protein aggregates, despite these patients often having symptoms quite different from the childhood disorder. Adult AxD patients typically display normal development, normal head size, and rarely have seizures. They do share hyperreflexia and pseudobulbar signs with the infantile cases, but more uniquely often have ataxia, visual and autonomic abnormalities, sleep disturbances and palatal myoclonus. For both the childhood and adult forms, no single symptom is present in all cases, and both may also display multiple other symptoms with lesser frequencies (Prust et al. 2011; Yoshida et al. 2011) . MRI diagnosis of adult cases is usually much more difficult than for children, since many of the criteria that apply for children do not hold for adults (van der Knaap et al. 2006) . In particular, instead of the frontal predominance of pathology seen in the childhood cases, the pathology in adult AxD is typically more caudal, involving the brain stem, cerebellum and cervical spinal cord. Importantly, white matter lesions are not found in about half of the adult onset cases (Yoshida et al. 2011) . Accordingly, AxD is now better considered as an astrogliopathy rather than a leukodystrophy.

Until recently, AxD was divided into three categories: infantile, with onset before 2 years of age; juvenile, with onset between 2 and 12 years; and adult, with onset at 13 years of age or later (Russo et al. 1976) . The infantile and adult onset cases usually differed from each other as noted above, whereas the juvenile cases could be like either of these, or display selected symptoms of each. Statistical analysis of 215 cases by Prust et al. (2011) found the data are better fit by a two class model rather than by three classes, Type I having an early onset and largely mirroring the infantile form, and Type 2 usually having a later onset and largely mirroring the adult form. Average age at onset for Type I was 1.7 years, and average lifespan following onset was 14 years; for Type 2 average onset was at 22 years, and average lifespan thereafter was 25 years. Here we adopt this Type I and Type 2 classification, which are also referred to as early onset and late onset.

Due to the lack of specificity of its clinical presentations, the late onset form of AxD was diagnosed much less frequently than the early onset form until the discovery that approximately 95 % of both are due to heterozygous missense mutations in the gene encoding glial fibrillary acidic protein (GFAP) , the primary and largely cell-specific intermediate filament protein in astrocytes (Brenner et al. 2001; Rodriguez et al. 2001; Gorospe et al. 2002; Li et al. 2005) . With the advent of this simple genetic diagnostic, adult cases are now discovered at a frequency similar to or even greater than early onset cases (Pareyson et al. 2008; Yoshida et al. 2011) , and include both de novo and familial occurrences. The cause of the remaining 5 % of cases for which a GFAP mutation has not been discovered is unknown. Since overexpression of wild type GFAP can produce AxD-like pathology in a mouse model (Messing et al. 1998) , the possibility of GFAP gene duplication has been examined for several patients without a detectable mutation, but with negative results (Yoshida and Nakagawa 2012) . The overall incidence of AxD has been estimated to be on the order of one in a million (Yoshida et al. 2011) .

The disease-causing mutations have been found throughout the length of the protein, with the exception of the non-helical N-terminal domain, and nearly all are 100 % penetrant. Genotype/phenotype correlations have been established for a few of the mutations and the clear absence of a correlation for several others, but the majority have been too rare for statistical analysis. As examples, mutations at R79 and R239 are associated with early onset and poor prognosis, whereas those at R88 and R416 have no correlation with age of onset or severity, suggesting other contributing factors to disease severity (Prust et al. 2011) . The GFAP mutations appear to act by a dominant, gain of function mechanism, since mice lacking GFAP have only mild deficits which do not resemble AxD (Gomi et al. 1995; Pekny et al. 1995; McCall et al. 1996) . This has implications for gene therapy, since provision of wild type GFAP to AxD patients is unlikely to be beneficial, and might instead exacerbate disease by increasing the GFAP load.

What critical function (or functions) of astrocytes is disrupted in AxD, and how GFAP mutations cause that disruption, are active areas of research, which is being pursued by analysis of human tissues, by cell free assays, study of primary and stable cell lines, and use of Drosophila and mouse models. All these studies involve comparing the effects of expressing wild type and mutant GFAP. Functions found altered include decreased proteasomal activity, decreased plasma membrane glutamate transport [through GLT-1 (rodents)/EAAT2(human), commonly named (as per HUGO gene nomenclature committee) solute carrier family 1 (glial high affinity glutamate transporter), member 2 (SLC1A2)], increased autophagy, activation of the stress-activated protein kinase/c-Jun N-terminal kinase (JNK) pathway; presence of advancedglycation and lipid peroxidation end products, and activation of an oxidative stress response and iron accumulation (Hagemann et al. 2006, 2009; Tang et al. 2006; Castellani et al. 1997, 1998; Tang et al. 2008; Tian et al. 2010; Wang et al. 2011; Sosunov et al. 2013) . In addition, AxD astrocytes are chronically reactive (Sosunov et al. 2013) , and thus likely to be secreting cytokines such as tumor necrosis factorα and interleukin1β, which can be toxic to other cells. Addressing any of these altered functions could potentially be therapeutic.

A mechanistic linking of GFAP mutations to proteasomal inhibition and stress pathway activation has been proposed by Tang et al. (2006) on the basis of cell culture studies. They found that expression of mutant GFAP inhibits proteasome function, that direct inhibition of proteasome function activates JNK, and that direct activation of JNK inhibits proteasome activity. This prompted the proposal of a positive feedback cycle in which mutant and/or increased GFAP inhibits proteasome activity, which in turn activates JNK, which further inhibits proteasome activity, resulting in yet higher GFAP levels. This is an attractive model because it links the primary defect in AxD, mutant GFAP, to physiological changes. It also could explain why mutations present from conception might not have consequence until later in life—effects might remain subthreshold until some central nervous system (CNS) perturbation increases GFAP above a critical level that sets the positive feedback loop in motion. However, what players in the loop contribute to the clinical manifestations of the disease remains unclear.

Another proposed mechanism linking GFAP mutations to the disease process is inactivation of one or more critical proteins through sequestration in the Rosenthal fibers, a mechanism that has been proposed for other neurodegenerative disorders (Ali et al. 2010) . Although GFAP is a major constituent of Rosenthal fibers, also associated with these aggregates are the small stress proteins HSP27 and αB-crystallin, the 20S proteasome subunit, plectin, p-JNK and p62 (Iwaki et al. 1993; Head et al. 2000; Zatloukal et al. 2002; Tang et al. 2006; Tian et al. 2006) . Proteomic studies of Rosenthal fibers are currently underway to identify other candidate proteins (Heaven and Brenner, manuscript in preparation) . Sequestration of the 20S proteasome has been suggested as a contributor to impaired proteasome activity in AxD, although as discussed below, it is also inhibited by a soluble GFAP oligomer (Tang et al. 2010) .

Both sequestration of αB-crystallin in the Rosenthal fibers and its ability to debundle GFAP filaments (Koyama and Goldman 1999; Perng et al. 1999; Nicholl and Quinlan 1994) led to it being tested therapeutically in a mouse model of AxD. This model consists of mice that both overexpress human GFAP from a transgene and have a knocked-in GFAP R236H mutation, which is the mouse homologue of the particularly pernicious human R239H mutation. Separately, the overexpression and the knock-in each display some characteristics of AxD, including formation of Rosenthal fibers, activation of stress responses and seizure susceptibility, but have normal lifespans. When combined together, however, they produce lethality between about 25 and 35 days of age, presumably from seizures (Hagemann et al. 2006) . In this lethal AxD mouse model, expression of αB-crystallin in astrocytes from a GFAP promoter-driven transgene to further increase the already elevated level present in the disease models resulted in complete rescue of viability, reduction of the stress response and partial restoration of GLT-1 levels (Hagemann et al. 2009) . These marked and general therapeutic effects thus recommend elevation of αB-crystallin for further investigation as an AxD treatment. Indeed, systemic delivery or increased expression of αB-crystallin may have general applicability beyond AxD, as it appears protective in models of neuroinflammatory disease and stroke (Ousman et al. 2007; Arac et al. 2011; Shao et al. 2013) . However, there are concerns about αB-crystallin as a therapy, primarily due to its identification as a negative risk factor in glioblastomas and several other types of cancer (Goplen et al. 2010) . Pharmacological strategies for enhancing αB-crystallin expression would not be tissue specific, and therefore enhancing αB-crystallin expression may be challenging to achieve without unacceptable side effects unless an astrocyte-specific delivery method could be developed, such as with viral vectors (see below).

Presumably the causes of the pathological changes in AxD astrocytes have as their origin a misfolded state of GFAP, but the identity of the toxic species, and whether it is the same immediate cause for each of the pathological changes, is not known. Correlations between the presence of Rosenthal fibers and disturbance of astrocyte functions have led to the tacit assumption amongst most investigators that Rosenthal fibers are the toxic species. However, by analogy with other protein aggregate neurodegenerative disorders such as Alzheimer’s disease, Parkinson’s disease and Huntington’s disease, the aggregate may be benign, and a smaller oligomer (which may be on the pathway to aggregate formation) may cause toxicity (Caughey and Lansbury 2003) . Support for this has been obtained by Tang et al. (2010) with regard to proteasome inhibition. They found that 20S proteasome activity was inhibited more severely by in vitro polymerized R239C mutant GFAP than by wild type GFAP, and that an oligomeric fraction was much more potent than the filament fraction. Native gel separation of the wild type oligomer fraction indicated that it contained primarily monomers and tetramers, whereas the R239C oligomers were primarily tetramers and a more slowly migrating species that could be an octamer. Exposure to αB-crystallin converted the R239C size distribution to that of the wild type, and also reduced proteasome inhibition, suggesting that the larger oligomer is a primary inhibitory species, and that a therapeutic activity of αB-crystallin is to prevent its formation. If the toxic species is indeed an oligomer rather than the Rosenthal fiber aggregate, the possibility exists that treatments targeting dissolution of the Rosenthal fibers could contrarily exacerbate disease by increasing oligomer levels.

Studies in Drosophila have produced indications of what may be primary and what may be secondary pathological effects in AxD. Although Drosophila have neither astrocytes nor GFAP , targeting mutant GFAP expression to Drosophila glia produced many of the phenotypes of AxD, including formation of Rosenthal fiber-like protein aggregates , loss of glia and neurons, susceptibility to seizures , and activation of stress pathways (Wang et al. 2011) . As in the mouse studies, over-expression of αB-crystallin decreased cell death, seizure susceptibility and aggregate formation, indicating that it acts at a fundamental level to reverse the AxD phenotype. On the other hand, although toxicity was decreased by elevating levels of catalase or superoxide dismutase to reduce oxidative stress, or by increasing plasma membrane glutamate transport, there was no effect on the burden of GFAP aggregates, suggesting that these treatments address downstream effects of the primary lesion. Also of therapeutic interest, the Drosophila studies found benefit from two compounds available for human use, vitamin E and 17-(allylamino)-17-demethoxygeldanamycin (17-AAG, an Hsp90 inhibitor that stimulates degradation of select proteins, and is undergoing clinical trials for cancer treatment).

Thus for treatment strategy, reducing GFAP levels by inhibition of synthesis or increased degradation is likely to be most efficacious, followed by prevention of misfolding by agents such as αB-crystallin, followed by treatment of downstream consequences such as activation of stress pathways and reduced plasma membrane glutamate transport. Both mouse and cell culture studies (Koyama and Goldman 1999; Mignot et al. 2007) (Messing unpublished observations) have found that Rosenthal fibers can disappear with time; thus to the extent that Rosenthal fibers serve as an indicator for the toxic GFAP species, limiting GFAP levels may not only prevent disease progression, but also reverse some symptoms. There is little concern about possible deleterious effects from extreme reduction in GFAP levels, since GFAP knockout mice have minimal neurological deficits. Tolerance of reduced GFAP expression in humans is suggested by the finding that 4 of 2026 healthy subjects were found heterozygous for deletion of the GFAP gene (Shaikh et al. 2009) .