Transcriptome analysis of microglia in a mouse model of Rett syndrome: differential expression of genes associated with microglia/macrophage activation and cellular stress

Rett syndrome (RTT) is characterized by a progressive loss of cognitive, social, and motor skills after a relatively brief period of typical development. It is found in ~1/10,000–1/15,000 female births and is usually due to de novo loss of function mutations in the X-linked gene, MeCP2, which codes for methyl-CpG binding protein 2 (MeCP2) [1]. MeCP2 binds to methylated cytosine at CpG islands, which are the canonical DNA methylation sites, as well as to non-CpG cytosine residues, and 5-hydroxymethylcytosine [2‐4]. It has also recently been found to inhibit microRNA processing [5]. Although severe loss of function mutations are usually male-lethal, hypomorphic MeCP2 variants have been found in males with intellectual disability and behavioral deficits [6, 7].

Rarely, mutations in CDKL5, SHANK3, FOXG1, ANKRD31, and CHRNA5 cause a RTT-like phenotype [8‐10].

One of the more perplexing aspects of RTT is the loss of previously acquired developmental milestones, which occurs after ~6–18 months of age. Regression is characterized by loss of language skills, reduced brain growth, repetitive stereotyped hand movements, and impaired motor skills [1, 2]. Following this period of regression, the clinical picture stabilizes for a while, but ultimately, motor deterioration, autistic features, seizures, growth failure, autonomic dysfunction, and gastrointestinal disturbances emerge.

In addition to RTT, de novo mutations in MeCP2 can contribute to schizophrenia (SZ) risk in a small subgroup of individuals [11]. And a recent genome wide association study (GWAS) carried out in a Han Chinese cohort suggests that common variants in MeCP2 might also play a role in this condition [12].

Although Mecp2 is ubiquitously expressed, most studies point to neuronal dysfunction as a primary cause. For example, a number of different neuron-specific Mecp2 KO mice show functional abnormalities [13‐15]. In addition, restoring Mecp2 expression in neurons normalizes brain weight and activity and extends lifespan [16]. An increase in cell packing density and a reduction in the complexity of neuronal dendritic branching have also been found, as well as alterations in dendritic spine numbers and synaptic architecture [17, 18]. Selective loss of Mecp2 in gamma-aminobutyric acid-ergic (GABAergic) inhibitory interneurons recapitulates most of the RTT phenotype [19]. However, some RTT features are also seen when selective loss of expression is induced in excitatory glutamatergic neurons [20]. In addition, the absence of Mecp2 has been found to cause a decrease in excitatory synapses and to impair long-term potentiation, an effect that was found in severely phenotypic, but not pre-phenotypic mice [21]. Interestingly, MeCP2 duplications also cause neurodevelopmental problems in humans and mice and are associated with increased synaptogenesis and dendritic complexity in vitro [2, 22, 23].

Although a primary neuronal dysfunction is clearly important in RTT, neurogenesis and synaptic function are modulated by other cell types in the brain—astrocytes and microglia, for example—and loss of function abnormalities in MECP2 in these cells could conceivably play a role in some aspects of the RTT phenotype. This idea is supported by several studies. For example, reduced Mecp2 expression in astrocytes negatively influences neuronal function, and reexpression improves locomotion and anxiety symptoms, restores respiratory abnormalities and dendritic morphology, and increases lifespan [24]. In addition, Mecp2-deficient astrocytes and conditioned medium from these cells failed to support normal dendritic morphology of wild-type (WT) and Mecp2-deficient hippocampal neurons. Also, mice with oligodendrocyte lineage-specific reduction of Mecp2 develop severe hindlimb clasping phenotypes, and restoration in an RTT model improved locomotor deficits and hindlimb clasping in both male and female mice, and restored body weight in males [25].

Finally, microglia have been implicated in RTT. Jin et al., for example, found that an increase in the expression of Slc38a1 (which codes for a glutamine transporter) caused by Mecp2 deficiency results in a glutamine-dependent decrease in microglia viability and a reduction in brain microglial number through the production of reactive oxygen species (ROS), which causes mitochondrial dysfunction and neurotoxicity [26]. In addition, a defect in microglial phagocytosis, which is a key homeostatic process in maintaining normal synaptic architecture, has been found in an RTT model [27]. Remarkably, phagocytic activity restored by a peripheral bone marrow transplant from WT mice, which led to the engraftment of phagocytic cells in the brain, rescued many aspects of the RTT phenotype in a mouse model; an increase in body weight and locomotor activity occurred, and breathing patterns were normalized [27]. Wang et al., however, were unable to replicate this finding, casting some doubt on the role of microglia in RTT [28]. In addition, Schafer et al. recently found that while microglia may contribute to the RTT phenotype by enhancing engulfment and elimination of presynaptic inputs at the end stages of disease, specific loss or gain of Mecp2 expression in microglia were not primary event [29].

To further address the potential role of microglia in RTT, we now report a transcriptome analysis in female Het mice. Transcriptomes were assessed during the transition from the RTT pre-phenotypic to the phenotypic stages (5 and 24 weeks, respectively). This time course models the natural progression of clinical RTT. In addition, our study is relevant in that, we used female mice; studies with mouse models of RTT rarely use females because symptoms occur earlier and are more severe in males, yet females account for nearly all RTT cases.

Microglia were isolated from 5-week-old Het female mice who did not exhibit overt neurological signs and 24-week Het female mice who exhibited a significantly increased hindlimb clasping reflex (p = 7.4E−08, Fig. 1). Microglia were also isolated at the same time points from age-matched female WT controls. Cells were immediately frozen at −80 °C. RNA was subsequently extracted and sequenced by a strand-specific protocol (RNA-seq). The 5-week samples (Het and WT) were sequenced together in a single lane by multiplexing. The 24-week samples were similarly handled. RNA-seq reads ranged from 49,135,314–98,691,394 in the week 5 samples to 25,463,864–31,288,101 in the week 24 samples (Additional file 2: Table S1). As seen in the file, the number of mapped reads exceeded 90% for all samples and all samples passed similar quality controls.

DEGs comparing WT and Het samples were identified using the DESeq2 software package and adjusted p values (padj) <0.05. A total of 464 DEGs were identified at 5 weeks; 78 increased in the Het samples, while 386 decreased (Additional file 3: Table S2). This was somewhat unexpected since Mecp2 is generally viewed as a transcriptional repressor, although it certainly can act as a transcriptional activator as well [41]. For the 24-week samples, 79 DEGs were found at the padj <0.05 level (42 increased in the Het samples, 37 decreased). The DEGs separated the WT (Mecp2 ^+/+) and Het samples (Mecp2 ^+/−) into two groups, as seen in the heat map (Fig. 2). In addition, Principle component analysis (PCA) using the top 5000 most variable genes, as defined by DESeq2 showed the samples could be separated, although one of the week 24 samples (Het2) showed increased variation from the other two biological replicates (Additional file 4: Figure S2).

It should be noted that the greater number of DEGs in the 5-week than 24-week samples is unlikely due to deeper sequencing depth since a similar number of week 5 DEGs were obtained using half of the reads for analysis. After randomly sampling reads from the 5-week samples five times to match the read depth of the week 24 samples, we identified between 27 and 33 up-regulated and between 224 and 257 down-regulated genes in the down-sized 5-week samples, with the same statistical thresholds (Additional file 3: Table S2). Although these numbers are smaller than those obtained with the full data set, they are still much larger than the total number of week 24 DGEs (n = 79), strongly suggesting that read depth is not a factor causing few DEGs in week 24 samples. Additionally, we examined the expression level of all week 5 DEGs in the week 24 samples and found that these genes exhibited similar levels of expression in the week 4 and week 24 samples (Additional file 5: Figure S3).

Among the top DEGs in the 24-week samples were Mecp2 (padj = 0.0001). In the 5-week samples, the decrease of Mecp2 mRNA was nominally significant (p = 0.002), with the adjusted p value falling just short of significance (0.058) (Additional file 3: Table S2).

In the 5-week samples, 24 of the top 25 DEGs were down-regulated in the Het samples. The top five were Sema3b, Nov, Dnaja1, Vcam1, and Mgp, three of which (Sema3b, Nov, and Vcam1) code for cell-cell adhesion proteins or extracellular matrix (ECM) components. This is supported by the gene list enrichment analysis using the ToppGene suite [35], which reported a significant enrichment of genes associated with the ECM among down-regulated genes, resulting in ECM being the top Biological Process GO term and cell adhesion being the third (Fig. 2; see Additional file 6: Table S3 for complete enrichment analysis).

Dnaja1 codes for a member of the heat shock protein (HSP) family of co-chaperones and is one of nine HSP-related genes that were significantly decreased in the 5-week Rett samples (Table 1). These findings were validated by qPCR (Fig. 3). According to the ToppGene enrichment analysis, genes coding for HSPs were the top protein domain detected among the down-regulated DEGs (Additional file 6: Table S3). In addition, qPCR analysis carried out on the CD11b-negative (non-microglia) fraction showed that at 5 weeks, Hspa1b was only one of the 5 HSP genes assayed that significantly decreased and a significant increase in HSPA8 gene expression was detected. By contrast, HSP genes were not differentially expressed in the 24-week samples in either the microglia or non-microglia fractions (Fig. 3; Additional file 6: Table S3).

The top up-regulated genes in the 5-week samples were C130026I21Rik, Cbln3, Gm10800, Retnla, and Rnf17. Of these, Retnla is the most interesting because it is a marker of M2 macrophage/microglial activation [42]. The M1 and M2 activation states regulate pro-inflammatory and anti-inflammatory balance, respectively, in the brain and immune system, dysregulation of which has been implicated in ASD in patients and animal models (see below) [42‐45].

There were fewer DEGs that showed an increase in expression in the 5-week samples, so the pathway analysis was less revealing. The top molecular function GO term was aminoacyl-tRNA ligase activity, which included the genes Iars, Yars, and Mars, one of which (Yars) has been implicated in developmental delay [46] (Additional file 3: Table S2). MHC class II was the top term based on the ToppGene analysis of PubMed literature, consistent with an effect on innate immunity (discussed below).

In the 24-week samples, the top protein-coding DEGs (up- or down-regulated) were Cass4, Ifit2, Nfkbid, Rian, and Lsp1 (Additional file 3: Table S2). Ifit2 codes for an interferon-inducible gene that inhibits virus replication [47]. Cass4 codes for a scaffolding protein that has been implicated in Alzheimer’s disease, possibly through the toxic effects of Tau [48, 49]. Nfkbid codes for an NF-κB inhibitor [50]. Interestingly, the blood level of NFKBID RNA has been proposed as a biomarker in the differential response children with ASD to have atypical antipsychotic medications [51]. RIAN (MEG8 in humans) is a non-coding RNA that maps to the human imprinted locus on 14q32. And Lsp1 (lymphocyte-specific protein 1) codes for an intracellular F-actin binding protein that regulates neutrophil chemotaxis [52]. It is one of the few DEGs found in both the 5- and 24-week samples (increased expression in the Het samples for both). In addition to Lsp1, Lpl (lipoprotein lipase) was also differentially expressed at both 5 and 24 weeks. However, although Lsp1 expression is lower in the Het samples at weeks 5 and 24, Lpl expression is higher in the Het samples at 5 weeks but lower at 24 weeks. Interestingly, reducing Lpl expression in BV2 microglia and primary microglial cells reduces microglial phagocytosis of fibrillar β-amyloid [53]. Correspondingly, reduced Lpl immunoreactivity has been found in granule cells of the dentate gyrus and the associated synaptic network in Alzheimer’s disease postmortem samples compared to control tissue [54]. The altered expression of Lpl in both sets of samples, and its differential expression between the 5- and 24-weeks samples, supports the idea that microglial phagocytosis could be disrupted in the RTT, as suggested in other studies [27, 55].

Because of the small number of DEGs, the ToppGene enrichment analysis for the 24-week DEGs should be viewed with caution. Nevertheless, several GO terms stand out. Notably, the top molecular function GO term enriched for DEGs that increased in the Het samples was glucocorticoid receptor binding, and abnormal innate immunity was the top mouse phenotype (Additional file 6: Table S3). This is consistent with a recent study showing that Mecp2 deficiency leads to dysregulation of inflammatory responses in microglia and macrophages and to abnormalities in genes induced by glucocorticoids [56].

Similarly, the top terms for co-expression genes enriched with DEGs were related to aspects of innate immunity, including genes down-regulated in macrophage cells stimulated with LPS, a TLR4 agonist, and genes regulated by NF-kB in response to TNF.

The role of microglia in ASD in general and RTT in particular is an intriguing pathophysiological perspective to consider. Several transcriptome studies show that microglia- and immune/inflammatory-associated genes are among the top DEGs in comparisons made between the ASD and control brains [45, 57, 58]. However, a key question is whether this activation is a primary or secondary phenomenon. The existing evidence points to the latter. In a study by Gupta et al., for example, a negative correlation was found between the differentially expressed neuronal module and the M2 activation state [45]. In addition, Voineagu et al. showed that the transcriptome changes they observed in the ASD brains converge with GWAS data, which are enriched for neuronal genes, in particular genes coding for synaptic proteins; immune changes have a less pronounced genetic component, suggesting that the differential expression found in ASD is secondary [58]. This idea is consistent with the key role of microglia in synaptic pruning [59‐61]. A reasonable hypothesis is that a primary defect in neurons leading to dysfunctional synaptogenesis is triggering microglial activation and synaptic pruning, which is then reflected in the postmortem gene expression findings. The importance of dysfunctional synaptic pruning in neuropsychiatric disorders was highlighted in a recent study showing that allelic variation in complement component 4 (C4) genes is associated with schizophrenia: C4 is a key regulator of microglia-mediated synaptic pruning [62]. In addition, genetic studies show that genetic variation in CSMD1, which codes for a complement regulator, is a risk factor for both ASD and SZ [63, 64].

On the other hand, there is some support for the idea that a primary defect in microglial function could also be involved in some ASD cases. For example, disruption of genes that regulate microglial phagocytosis, such as annexin 1 (ANXA1) and MER proto-oncogene tyrosine kinase (MERTK), which are expressed at markedly elevated levels in microglia compared with neurons [65], have been found in rare, ASD-related CNVs and loss of function mutations in exome sequencing studies [66‐68].

The possibility for a direct role of microglia in RTT is controversial, however. As noted in the introduction, restoring macrophage function in a mouse RTT model was reported to alleviate some symptoms in one study, but not another [27, 28]. In addition, Schafer et al. suggested recently that while microglia may contribute to the RTT phenotype by enhancing engulfment and the elimination of presynaptic inputs at the end stage of disease, specific loss or gain of Mecp2 expression in microglia were not primary events [29]. On the other hand, Mecp2 deficiency leads to dysregulation of inflammatory responses in microglia and macrophages, which could alter their function and contribute to some aspects of disease pathogenesis [56], which is supported by our findings .

With these findings in mind, we carried out a transcriptome study in microglia derived from female Het mice, a model that more closely mimics clinical RTT, which occurs in females. One of our significant findings was that genes coding for nine HSP proteins were markedly decreased in microglia from pre-phenotypic female RTT mice. By contrast, no differences in HSP gene expression were detected in the 24-week samples derived from mice that had developed neurological symptoms. This suggests that in pre-phenotypic female mice, microglia could be susceptible to hyperthermia and perhaps other cellular stressors that function through the HSP pathway, such as oxidative stress, hypoxic stress, and inflammation. HSPs are molecular chaperones that play a role in brain function by preventing protein misfolding and by promoting the degradation of proteins that are improperly folded [69, 70]. Upon exposure to cellular stress, an increase in expression occurs, which protects cells from apoptosis. Although there are no studies suggesting that HSPs play a role in RTT, there is some evidence pointing to other forms of cellular stress having an impact. For example, reduced Mecp2 expression leads to redox imbalance and oxidative stress [71‐73]. In addition, increased expression of hypoxia-induced transcripts was found in Mecp2-deficient microglia and peritoneal macrophages [56].

Another significant finding in our study is that both 5- and 24-week DEGs were significantly enriched for genes that are associated with both the M1 and M2 activation states, suggesting that neuroinflammation is an active process in both pre- and post-phenotypic RTT mice. Curiously, however, disparate M1 and M2 gene sets were differentially expressed in pre-phenotypic and phenotypic microglia. This could be due, perhaps, to a temporal difference in microglial activation in RTT that are specific to microglia per se, or exogenous, due to interactions with other cell types in the brain; microglia can be activated through contact with neurons via the pruning of weak, ineffective synapses, and with activated astrocytes, oligodendrocytes, and endothelial cells [74‐76].

Disparate findings in pre-phenotypic and phenotypic microglia were also observed by Cronk et al. who showed that compared with control males, microglia have significantly smaller soma in pre-phenotypic Mecp2-null mice, but larger soma (indicative of microglial activation) in phenotypic mice [56].

Although overall, there was overlap with the Cronk et al. transcriptome findings on a general level (i.e., altered expression of inflammatory and cellular stress genes); there was limited overlap in DEGs per se, perhaps owing to differences in sex; we used females and their RNA-seq data were obtained using males, or temporal differences (early phenotype vs late phenotype).

Finally, the enrichment of DEGs that contribute to the ECM in the 5-week RTT samples is interesting in view of the contribution made by microglia, albeit modest compared with astrocytes and neurons, to ECM homeostasis [77‐81]. Considering the importance of the ECM in maintaining synaptic function, the altered expression of ECM components and regulators in the pre-phenotypic phase could play some role in the synaptic dysfunction that ultimately leads to clinical symptoms.

A caveat to our findings is that the microglia from Het mice were not separated into WT and Mecp2-null cells. Thus, X chromosome skewing could have had an impact on our results. We computed the ratio of RNA-seq reads mapped to the deleted exons (exons 3 and 4) vs the reads mapped to the undeleted exons (exons 1 and 2) and then compared the ratios between Mecp2 Het and WT samples. Based on this, we estimate that the Mecp2-null cells were probably present in ~40% of the cells used for RNA sequencing (data not shown) suggesting that X chromosome skewing did not have a significant impact on our findings. However, follow-up single-cell sequencing studies will be necessary to better resolve this.

In addition, follow-up functional studies are needed to validate our findings, in particular, whether microglia derived from pre-phenotypic RTT mice show an altered response to heat shock and other stressors that require an intact HSP response, the dysfunction of which might compromise microglial function, contributing to the development of symptoms later in life.

Transcriptome analysis was carried out on microglia from control female mice and heterozygous female mice carrying one Mecp2-null allele—a mouse model of Rett syndrome—at two time points; 5 weeks, prior to the onset of neurological symptoms (pre-phenotypic) and 24 weeks, after the development of neurological symptoms (phenotypic). Genes involved in innate immunity and in M1 and M2 macrophage activation were differentially expressed at both time points, although different sets of genes were found at the two time points. In addition, a number of heat shock genes were differentially expressed at 5 weeks, but not at 24 weeks. These findings suggest that pre-phenotypic female mice may have alterations in their capacity to response to heat stress and other stressors that function through the HSP pathway, as well as dysregulated expression of genes involved in innate immunity, both of which might contribute to the later development of neurological symptoms.